Exaptation: Academic mentees’ career pathway to be independent and impactful

Yanmeng Xing Department of Statistics and Data Science, Southern University of Science and Technology, Shenzhen 518055, Guangdong, China Ye Sun Tongxin Pan Department of Statistics and Data Science, Southern University of Science and Technology, Shenzhen 518055, Guangdong, China Xianglong Liang Department of Statistics and Data Science, Southern University of Science and Technology, Shenzhen 518055, Guangdong, China Giacomo Livan Dipartimento di Fisica, Universit‘a di Pavia, via Bassi 6, 27100, Pavia, Italy Department of Computer Science, University College London, 66-72 Gower Street, WC1A 6EA London, UK Yifang Ma

Abstract

In science, mentees often follow their mentors’ career paths, but exceptional mentees frequently break from this routine, sometimes even outperforming their mentors. However, the pathways to independence for these excellent mentees and their interactions with mentors remain unclear. We analyzed the careers of over 500,000 mentees in Chemistry, Neuroscience, and Physics over the past 60 years to examine the strategies mentees employ in selecting research topics relative to their mentors, how these strategies evolve, and their resulting impact. Utilizing co-citation network analysis and a topic-specific impact allocation algorithm, we mapped the topic territory for each mentor-mentee pair and quantified their academic impact accrued within the topic. Our findings reveal mentees tend to engage with their mentors’ less-dominated topics and explore new topics at the same time, and through this exaptive process, they begin to progressively establish their own research territories. This trend is particularly pronounced among those who outperform their mentors. Moreover, we identified an inverted U-shaped curve between the extent of topic divergence and the mentees’ long-term impact, suggesting a moderate divergence from the mentors’ research focus optimizes the mentees’ academic impact. Finally, along the path to independence, increased coauthorship with mentors impedes the mentees’ impact, whereas extending their collaboration networks with the mentors’ former collaborators proves beneficial. These findings fill a crucial gap in understanding how mentees’ research topic selection strategies affect academic success and offer valuable guidance for early-career researchers on pursuing independent research paths.

Introduction

For young scientists, mentorship is widely regarded as a cornerstone in shaping their academic career trajectories[1, 2, 3, 4, 5, 6, 7]. Through close mentorship, mentees often follow in the footsteps of their mentors, adopting similar research interests and methodologies. This alignment can provide a strong foundation for early career success, as mentees benefit from their mentors’ expertise, networks, and established reputations[8, 9, 10]. However, the very essence of mentorship goes beyond mere imitation; it involves guiding mentees to become independent thinkers who can eventually contribute to their fields in novel and impactful ways.

A compelling illustration of mentorship’s transformative power is found in the story of Giuseppe Levi, an anatomist whose mentorship produced three Nobel laureates in Physiology or Medicine: Salvador Luria, Renato Dulbecco, and Rita Levi-Montalcini[11]. These mentees not only followed their mentor’s path but also expanded beyond it, ultimately achieving greater recognition than their mentor. Such cases highlight the potential for mentees to surpass their mentors, raising a crucial question: How can early-career researchers strategically select their research topics to rapidly establish independence and achieve significant academic impact? How to interplay with the mentor’s research territory to avoid competition and reach a win-win situation? The challenge of selecting a research direction is particularly pronounced for newly independent principal investigators (PIs)[12]. Continuing with their mentor’s research line may lead to quick publications, but it risks creating competitive tensions. Conversely, pioneering new directions involves high risks and a substantial trial-and-error cost[13]. This dilemma underscores the importance of understanding how mentees navigate their research topic selection and the subsequent impact on their academic careers.

Firstly, the strong bond of mentorship makes it difficult to establish a totally new research direction out of the mentor’s territory, which is demonstrated by the profound impact of mentorship on mentees’ success across various dimensions, including enhancing research productivity, nurturing future scholars, and improving academic survival rates[8, 14, 10, 15, 16, 17, 18]. Early studies focused on the quantitative benefits of mentorship, such as the development of research skills, expansion of professional networks, and increased academic productivity [3, 19]. Effective mentorship also involves attending to mentees’ psychosocial needs, including advising on resource negotiation, and helping to balance work and life [5, 20]. These aspects of mentorship are crucial for promoting career support and persistence[21, 22], preventing burnout [14, 23, 24], improving research dissemination [2], increasing grant acquisition rates [25], and enhancing overall career satisfaction [26, 27, 28]. Secondly, with the increasing focus on scientific productivity[29], faster turnaround times, and the challenging funding achievement circumstance[30, 31], effective mentorship has become increasingly difficult, particularly in larger mentoring groups where unequal mentor engagement can negatively impact mentees’ survival rates in academia [16]. Furthermore, some mentors may restrict their mentees’ choice of research areas or project rights, particularly when mentees are not allowed to "port" projects from their mentors’ labs, which is often crucial for the success of junior scientists [32]. The specific dynamics of mentor-mentee collaboration can also become a double-edged sword. While close collaboration can provide significant support, it may also create dependency, limiting mentees’ academic horizons and innovation [33, 34, 28, 35]. This over-reliance on mentors can hinder mentees’ transformation into independent and distinguished researchers [33, 36, 37, 38]. At last, in the context of research topic selection and academic innovation[39, 40], the alignment between mentor and mentee has garnered considerable attention. Early-career alignment of research interests can foster productive collaborations and lead to higher publication outputs [33, 36, 41]. However, excessive involvement in a mentor’s research area may inadvertently stifle the development of a mentee’s independent research identity [37, 38]. Studies in STEM fields have shown that mentees who establish independence from their mentors’ research topics after graduation are more likely to succeed in academia [10, 35]. Additionally, interdisciplinary mentorship can encourage mentees to engage with diverse research questions, fostering research innovation and independence [42, 43, 44, 45]. In contrast, evidence suggests that frequently switching topics[13] or exploring new areas[46], particularly in the early stages of a career, correlates with lower academic impact.

In light of this dilemma for young scientists, this paper examines how mentees choose research topics in relation to their mentors’, how these strategies evolve over careers, and the resulting impact. We hypothesize that the transmission and development of scientific skills and knowledge within mentorships follow a process similar to "Exaptation", a concept from evolutionary biology[47, 48]. Exaptation describes a feature that originally evolves for one function but is later repurposed for another. Likewise, in mentorships, mentees initially immerse themselves in their mentors’ topics and later repurpose this knowledge to explore and develop their own novel research directions. Through this process, they achieve independence and build successful academic careers.

To validate this hypothesis, we analyze a longitudinal dataset encompassing mentorship and publication records in Chemistry, Neuroscience, and Physics from 1960 to 2021, sourced from the Academic Family Tree[49] and OpenAlex (Methods). By constructing co-citation networks for each mentor-mentee pair [50, 51]and applying community detection methods[52, 53], we identify the research topics they are engaged in. This allows us to examine how mentees balance leveraging their mentor’s expertise while cultivating their own research identities. We highlight the differences between the strategies employed by elite mentees and those of the general population. In addition, we explore the relationship between the degree of topic divergence from mentors and the long-term impact of mentees, finding an inverted U-shaped correlation. We also examine the influence of collaboration, discovering that excessive co-authorship with mentors can impede a mentee’s impact, while expanding their collaboration networks through their mentors’ former collaborators is beneficial. These findings illuminate how research topic selection strategies correlate with academic success and offer key insights for early-career researchers seeking independence.

Results

Co-citation network and topic-specific impact measure

Refer to caption — Figure 1: Illustration of a mentor-mentee co-citation network and topic-specific impact measurement. a. The co-citation network contains all the papers published by a mentor-mentee pair. Each node represents a paper: circles for mentee papers, triangles for mentor papers, and overlapping shapes for mentor-mentee collaborated papers. Links connect nodes if the corresponding papers are co-cited by at least one common paper. Node colors denote the different communities identified by the fast unfolding algorithm[53], and node sizes correspond to their impact, as determined by the algorithm described in panel c. b. The time series tracks the network’s evolution, depicting the growth of papers by the focal mentee (upper half) and the pair-wise mentor (lower half) throughout their careers. Each shaded area’s color matches a topic community in the co-citation network, with the height at each year point indicating the number of papers published on that topic. The vertical black line represents the year the mentor started supervising the mentee. c. The heuristic algorithm quantifies the impact of a focal mentee by analyzing co-citations within each topic community. The mentors’ papers ( $r_{j,i}$ ) and mentees’ papers ( $e_{j,i}$ ) are organized by their topic communities in the left column, color-coded to align with the communities $j$ identified in panel a. Squares in the right column signify papers that co-cite any of the mentor-mentee papers, with colors matching the communities of mentor-mentee papers. Solid lines connect papers with the same colors, showing co-citations within communities, while dashed lines indicate cross-community co-citations.

In this study, we analyze genealogical data from $0.5$ million mentorship records involving $80$ k scientists, each of whom published at least 20 articles, totaling nearly $10$ million papers in Chemistry, Neuroscience and Physics between $1960$ and $2021$ (Methods and Supplementary Table S2). The selection of these fields is justified in Supplementary Section 1.2.1 (Supplementary Table S1 and Fig. S1). The citation profiles of mentors’ and mentees’ papers, as well as the papers citing them, are extracted from the OpenAlex database. To examine how the papers of a mentor-mentee pair are related, we construct for each mentorship pair a co-citation network (Fig. 1a). In this network, each node represents a paper authored by either the mentor, the mentee, or both, and two nodes are connected if their corresponding papers are cited together by at least one common source. This method of linking nodes is widely used to detect research topic similarity, as papers co-cited by common sources are more likely to be semantically related than randomly paired papers[54, 55]. It does not escape us that our approach to building co-citation networks could be enhanced via a statistical validation method aimed at distinguishing statistically significant co-citations from co-citations that could effectively be considered random events [56, 57]. The downside of such an approach, however, would be that of ending up with exceedingly sparse co-citation networks that would not be amenable to any meaningful analysis. At any rate, as we detail next, we find our co-citation networks to be made of communities that can be meaningfully associated with different research topics (Supplementary Fig. S3).

The topic communities in each co-citation network are identified with the fast unfolding algorithm[53], a popular approach for detecting community structures in large networks[58, 59]. Typically, a co-citation network consists of several large topic communities, along with some smaller clusters and isolated nodes. To ensure the identified research topics are meaningful[52, 13, 60], we include only topic communities with at least $10$ nodes in this study. Following this procedure, the co-citation network also retains the majority of papers with citations from both the mentor and mentee (Supplementary Fig. S2). Fig. 1a illustrates a co-citation network of a typical mentor-mentee pair, featuring $6$ distinct topic communities, with each node coloured according to its identified topic community. Nodes are shaped as circles, triangles or triangles inside circles, representing papers authored by the mentee, mentor, or both, respectively. We have further validated the accuracy of the topic community detection algorithm in categorizing research topics by referring to the concept classification in OpenAlex, confirming that the majority of papers within the same community indeed belong to the same research topic (Supplementary Fig. S3). Fig. 1b illustrates the evolution of the research focus for a mentor-mentee pair, highlighting their respective publications on different topics throughout their careers.

To measure the research impact of a mentor-mentee pair, we adapt and apply a collective impact allocation algorithm[50, 61] to capture their respective contributions to the papers within each topic community. Fig. 1c presents an example of how we calculate the impact of a focal mentee with papers belonging to different detected topic communities. We first identify all the papers published by the mentee within a community $j$ , forming a set $E_{j}\equiv\{e_{j,1},e_{j,2},\ldots,e_{j,n_{j}}\}$ , and all the papers published by the mentor, forming a set $R_{j}\equiv\{r_{j,1},r_{j,2},\ldots,r_{j,m_{j}}\}$ . Here, $n_{j}$ and $m_{j}$ respectively represent the number of papers authored by the mentee and mentor in the community $j$ . The complete set of all papers published by the mentee and mentor in community $j$ is $M_{j}=E_{j}\cup R_{j}$ . We then identify all co-citing papers $P_{j}$ , which comprises the complete set of papers co-citing at least two papers together from the set $M_{j}$ . Based on this framework, the total impact $C_{j}^{e}$ accrued by a mentee from all their papers $E_{j}$ within community $j$ is defined as:

\displaystyle C_{j}^{e}=\sum_{i=1}^{n_{j}}\frac{w_{e_{j,i}}}{s_{e_{j,i}}}

(1)

where the impact $w_{e_{j,i}}$ is the number of citations that paper $e_{j,i}$ has received from the co-citing papers $P_{j}$ . The term $s_{e_{j,i}}$ represents the number of authors on the mentee’s paper $e_{j,i}$ , which serves to normalize the impact $w_{e_{j,i}}$ and fairly distribute the impact among co-authors[62]. Similarly, the total impact $C_{j}^{r}$ attained by a mentor from all their papers $R_{j}$ within community $j$ can be expressed as:

\displaystyle C^{r}_{j}=\sum_{i=1}^{m_{j}}\frac{w_{r_{j,i}}}{s_{r_{j,i}}}

(2)

where $w_{r_{j,i}}$ signifies the number of citations received by paper $r_{j,i}$ of the mentor from the co-citing papers $P_{j}$ , while $s_{r_{j,i}}$ represents the number of authors on paper $e_{j,i}$ . Through this framework, we can effectively quantify the scientific impact of each mentor-mentee pair within each of their topic communities.

Elite mentees strike a balance between research on their mentors’ topics and exploring new ones.

Once the detected communities are delineated in the co-citation network, an examination of the research topic differences between mentees and their mentors can be conducted. We classify the mentees’ strategies for selecting research topics into three distinct patterns (Fig. 2a and Supplementary Fig. S5a): (i) pure follow, wherein mentees publish papers solely on their mentors’ research topics; (ii) follow and innovate, wherein mentees work on their mentors’ research topics while also initiating new ones not explored by their mentors; (iii) pure innovate, wherein mentees venture exclusively into new topics independent of their mentors. We further define the degree of a mentee’s engagement in new topics, $R$ , as the ratio between the number of new topics the mentee undertakes and the number of all of their topics (Fig. 2a). The values of $R$ range from $0$ and $1$ , indicating from a pure follow strategy ( $R=0$ ) to a pure innovate strategy ( $R=1$ ). For instance, in the center of Fig. 2a, the mentee has published papers on three common topics with the mentor and explored one new topic, resulting in $R=1/4$ .

Fig. 2b-d insets illustrate the distribution of mentees under each strategy and compare it to the strategy choices of the elite mentees, who are defined as those whose total number of citations of papers over their careers ranks within the top 20% of all mentees. Here we only consider the number of citations a paper receives within the first five years after its publication, thereby reducing the cumulative citation effect over time[63]. This metric allows for a more equitable comparison of the early impact of papers published at different times, ensuring that elite mentees are represented across various eras (Supplementary Fig. S4a-c). Notably, both general and elite mentees are required to have published their first papers before 1990, guaranteeing that their academic careers can span over 30 years. This criterion enables us to examine the landscape of mentor-mentee topic engagement throughout their entire career. We observe a consistent strategy pattern across the three research fields of Chemistry, Physics, and Neuroscience. A larger fraction of all mentees adheres to the pure follow strategy compared to elite mentees. The majority of elite mentees adopt the follow and innovate strategy, with the orange bar significantly exceeding the blue bar, indicating that elite mentees are more likely to contribute to both shared and new topics. Although both all mentees and elite mentees have the smallest fractions in the pure innovate category, elite mentees are slightly less represented here than the general mentee population. Additionally, we examine the complementary cumulative distribution function (CCDF) of $R$ , which represents the probability that the degree of a mentee’s engagement in new topics is larger than $R$ (Fig. 2b-d). We observe that across three fields, there is one point where the probabilities of the elite mentee and the general mentee population are equal and the two distributions intersect. This further reveals that, compared to the general mentees, the majority of elite mentees prefer a balanced approach, combining their mentors’ topics with new and independent ones (follow and innovate, Supplementary Fig. S5b).

To further understand how mentees engage with their mentors’ research topics, we subdivide the mentors’ topics into primary and secondary categories. A topic is designated as "primary" if the proportion of the mentor’s papers published in that topic exceeds the median proportion across all their topics; otherwise, it is designated as "secondary". We have validated that mentors are significantly more involved in these primary topics than in secondary topics, suggesting the effectiveness of this dichotomy(Supplementary Fig. S6a,e,i). Note that both primary and secondary topics may encompass multiple distinct topics (Supplementary Fig. S6). We then examine how mentees distribute their research efforts across mentors’ primary topics and secondary topics, as well as their new topic communities in the three fields, as shown in Fig. 2e-g. Our analysis reveals that compared to the general mentee population, elite mentees in all three fields exhibit a consistently higher level of diversification under each topic type, particularly in secondary and new topics. This indicates that elite mentees tend to diversify their research efforts more extensively. In Chemistry and Physics (Fig. 2e and 2g), elite mentees show a particularly strong tendency to engage in new topics, suggesting a higher propensity for innovation and exploration beyond their mentors’ established areas. Additionally, we find that for mentees who have followed their mentors’ research topics, both elite and general mentees choose to engage less with their mentors’ primary topics rather than secondary topics.

Given the above findings on mentees’ engagement in three topic types, we are interested in determining which topic selection types yield the most scientific impact for mentees. To address this, we calculate the average ratio of scientific impact attributed to the mentees’ publications within each topic type. Fig. 2h-j presents the comparative results for both the general mentee population and elite mentees, categorized by their follow and innovate and pure follow strategies. We can see that both general and elite mentees, regardless of whether they choose a follow and innovate or pure follow strategy, achieve the lowest impact ratio when engaging in their mentors’ primary topics, and this pattern is consistent across all three fields. This is likely in part due to the fact that the saturation and established nature of mentors’ primary topics leave limited space for mentees to stand out and make significant contributions. Furthermore, elite researchers who adopt the follow and innovate strategy exhibit an interesting balance: they achieve a similar impact ratio on both new topics and their mentor’s secondary topics. This suggests their ability to effectively innovate while still leveraging their mentor’s influence in secondary areas. In contrast, the general mentee population shows a higher impact ratio in new topics compared to their mentors’ secondary topics. We further analyze the probability of mentees (both general and elite) publishing their highest impactful paper under each topic type (Supplementary Fig. S7), arriving at similar conclusions.

Mentees begin their careers by engaging with their mentors’ topics and progressively establish their own research domains.

We now proceed to extend this analysis by evaluating how mentees accumulate their scientific impact in each topic type over careers, and examining whether there are strategy differences between the general mentee population (all mentees) and those exceptional mentees who not only rank in the top 20% for citation impact among all mentees but also ultimately outperform the total impact of their mentors (outperforming mentees). Fig. 3a-c shows the average cumulative scientific impact of all mentees (dashed black line) and their mentors (solid black line) as a function of their career ages in the fields of Chemistry, Neuroscience and Physics. The results indicate that in Neuroscience and Physics, mentors consistently exhibit higher cumulative impacts compared to their mentees at the same career stages, while the mentees and their mentors in Chemistry show indistinguishable cumulative impacts at equivalent career ages. We then shift our focus on how mentees accumulate scientific impact within each topic type throughout their careers, represented by orange lines for primary topics, green for secondary, and grey for new topics. We can see that, the general mentee population accumulates a similar amount of impact in their mentors’ primary and secondary topics during the early-career stage (0-10 career years). As time progresses, the impact from secondary topics significantly overtakes that from primary topics after the mid-career stage (10-20 career years), suggesting a strategic shift towards secondary areas or the natural exhaustion of certain topics over time. This trend continues, with the gap between primary and secondary topics widening until the end of their careers. The mentee’s cumulative impact on new topics (grey lines) gradually grows throughout their career. While it starts at the bottom during the early stage, it eventually surpasses the impact of primary topics in the later stage (post-20 career years). On the contrary, the corresponding mentors tend to accrue the most citation impact in their primary research topics, rather than in secondary topics (Supplementary Fig. S8). The inset boxplots in these panels offer a detailed decade-by-decade breakdown of publication distributions by topic type across all three fields.

In contrast, outperforming mentees who exceed their mentor’s impact, as shown in Fig. 3d-f, begin their careers with a strong focus on secondary topics. We can see the green lines rise sharply, indicating these mentees’ tendency to quickly establish significant contributions to their mentors’ secondary topics. This early divergence from their mentors’ primary research interests is evident across Chemistry, Neuroscience, and Physics. During the early stages of their careers, these outperforming mentees initially accumulate nearly equal citations from both their mentors’ primary topics and their own new topics. However, they later receive more impact from new topics than from their mentors’ primary topics, achieving this shift at an earlier career stage compared to the general researcher population. Fig. 3 collectively illustrates the evolution of academic impact through strategic topic selection across disciplines. The successful career trajectories of these outperforming researchers highlight the potential rewards of a balanced and innovative research approach, providing a clear roadmap for aspiring academic success in science.

Mentees often outperform mentors in impact within their moderately similar research topics.

What kind of topic selection enables mentees to outperform their mentors? To answer this question, we first categorize the mentees into four quadrants based on whether their citation impact in primary or secondary research topics exceeded that of their mentors (Fig. 4a-c). We can see that the general mentee population (depicted in purple) shows a significant concentration in the third quadrant, with 60% for Chemistry, 61% for Neuroscience, 56% for Physics, indicating that the majority of mentees do not outperform their mentors in either primary or secondary topics over their careers. Followed by the proportion of mentees in the second quadrant, approximately 37% of mentees achieve more citation impact in secondary topics but less impact in primary topics compared to their mentors. A smaller proportion of mentees, specifically 2% for Chemistry and Neuroscience and 5% for Physics, are found in the first quadrant, where they outperformed their mentors in both primary and secondary topics. Lastly, only 1% of mentees across these three fields are located in the fourth quadrant, with more impact in primary topics and less in secondary topics, which highlights the difficulty of exceeding their mentor’s impact in the mentors’ established fields. Further examination of the distribution of outperforming mentees across these four quadrants reveals different strategic propensities for topic selection. Compared to the general mentee population, outperforming mentees (represented by yellow points) are more likely to be in the second quadrant, exceeding their mentors’ impact in secondary topics while lagging in primary topics. Taken together, the general mentee population navigates the challenging terrain of mentor-established primary topics, while outperforming mentees show a strong preference for secondary topics, marking a path of innovation and leadership. This trend is consistently observed across the disciplines of Chemistry, Neuroscience, and Physics.

To intuitively understand how mentees engage with different topic types, we plot each mentee on a triangle (Fig. 4d-f). A mentee is positioned at the top corner if all their citations originate from primary topics. The bottom-left corner represents exclusive impact in secondary topics, while the bottom-right corner indicates impact received only from new topics. When a mentee receives citations from multiple topic types, their position within the triangle reflects the mix of citations. For instance, a mentee would be in the center if they received an equal number of citations from all three topic types. For the general mentee population, we observe that while some of the mentees cluster near the corners, signifying a strong focus on a single type of topic, many others are evenly spread out, reflecting a balanced approach across primary, secondary, and new topics. In contrast, the outperforming mentees, marked by yellow circles, tend to spread along the secondary-new and primary-secondary edges, rather than in the middle of the triangle. This pattern further confirms that the general mentee population demonstrates a commitment to breadth in their academic pursuits, initially contributing to their mentor’s primary topics and subsequently expanding to other areas (Fig. 3a-c), while outperforming mentees prefer to distinguish themselves through significant advances in mentor’s secondary and their new topics (Fig. 3d-f). Also, we note that very few elite mentees are positioned along the primary-new edge of the triangle. This scarcity underscores the inherent progression in the academic journey, highlighting that directly transitioning from established primary topics to new ones is relatively rare and challenging.

How does the thematic similarity between a mentee’s work and that of their mentor influence the mentee’s ability to achieve high impact or outperform the mentor’s impact? To answer this, we quantify the topic distance between a pair of mentee and mentor by calculating the average length ( $L$ , formula 4) of the shortest path linking their papers within the co-citation networks. A greater topic distance indicates higher dissimilarity between the mentor’s and mentee’s research topics. Fig. 4g-i shows the average cumulative citations of mentees as a function of the topic distance between the mentee and their mentor. Across the disciplines of Chemistry, Neuroscience, and Physics, we observe consistent and significant inverted-U curves fitted by all of the data points (representing the general population of mentees) from the inset plots, supported by statistical significance tests (Supplementary Table S3-S5). This result confirms the hypothesis that there is an optimal balance between leveraging a mentor’s expertise and forging an independent research path, a balance that enables mentees to maximize their impact.

Effective use of mentors’ collaboration networks supports mentees’ development.

To further validate the inverted U-shaped relationship between the average network distance and the mentee’s cumulative impact (Fig. 4g-i), we performed a non-linear regression analysis that accounted for a set of potential confounding factors. The regression models were adjusted sequentially by incorporating control variables in the following order: mentee-related factors, mentor-related factors, and mentee-mentor collaboration factors (Supplementary Note S3.1). Supplementary Tables S6-S8 show that the R-squared value increased with the addition of control variables, indicating an improved fit. The analysis was limited to mentees with career lengths of at least 30 years, as they provide a comprehensive perspective on the dynamics of research divergence and cumulative career achievements. This approach ensures that the mentees with full career spans better represent the evolution of research interests over time and their total career outcomes.

The regression results incorporating all variables for three fields are reported in Fig. 5 (Supplementary Table S6-S8, Model 10). We observed that the linear effect of average network distance (ave_distance) on the mentee’s cumulative impact is significantly positive, and the quadratic term ( $ave\_distance^{2}$ ) exhibits a significant negative association with the mentee’s impact. These results provide statistical evidence supporting the inverted U-shaped relationship depicted in Fig. 4g-i. Moreover, the length of mentees’ careers (career_len_mte), the number of publications in their first five years of career (mte_work_count_first_5y), and the mentor’s citation impact are positively correlated with the mentee’s cumulative impact. This suggests that the mentee’s individual efforts and the mentor’s reputation significantly contribute to the mentee’s career development[64, 65]. In contrast, a higher number of research topics pursued by the mentor (topic_num_mto) negatively correlates with the mentee’s impact. This may be due to the mentor’s overly diversified interests, which could dilute the interaction and guidance provided to the mentee[16].

Intriguingly, we discover heterogeneity in the effects of the mentor-mentee collaboration on the mentee’s scientific impact. As indicated in Fig. 5b (Supplementary Table S7, Model 7), a significant negative relationship exists between the total count of mentor-mentee collaborative works (colla_work_count) and the mentee’s cumulative impact in Neuroscience. However, a differentiation emerges when we divide the collaboration into those within the first five years of the mentee’s publications (colla_work_count_first_5y) and those beyond the first five years (colla_work_count_later). Model 9 in Supplementary Table S7 reveals that colla_work_count_first_5y is positively associated with the mentee’s cumulative impact, whereas colla_work_count_later negatively impacts it, which is also consistent in Chemistry (Supplementary Table S6). This suggests that close collaboration with a mentor is beneficial for mentees in the early, formative stages of their careers, as it provides essential guidance and foundational knowledge[66, 64]. However, this positive effect tends to reverse if the reliance extends into the later stages of their careers. Therefore, the path to becoming a distinguished and independent researcher involves an initial phase of mentor-guided collaboration followed by a strategic shift towards self-directed research endeavors.

Supplementary Fig. S8 illustrates that mentees in Physics tend to collaborate more closely with their mentors during the first five years of their careers compared to those in Chemistry and Neuroscience, indicating a greater reliance on their mentors’ guidance in the early stages. Consequently, the variable colla_work_count_first_5y exhibits a negative effect in Physics, as shown in Fig. 5c. These findings highlight the importance of mentees developing independence and reducing reliance on their mentors to become successful researchers. Importantly, when we incorporate the variable of the number of common collaborators between mentor and mentee (common_cllaborators_count), Fig. 5a-c indicate that it plays a significant positive role in promoting the mentee’s career impact across three fields. This result suggests that while relying too much on a mentor can impede a mentee’s long-term impact, mentees can strategically leverage their mentor’s social connections to access broader resources and thereby accelerate their professional development.

Discussion, limitations, and conclusions

This study systematically contrasts the academic trajectories of all researchers with those of elite researchers, identified by the top 20% citation impact in their respective fields, revealing nuanced insights into the role of mentorship in shaping mentees’ academic development. Our findings indicate that both the general population of mentees and elite mentees are more likely to achieve greater success in their mentors’ secondary research topics and in pioneering new research domains, rather than in their mentors’ primary research areas. Notably, elite mentees tend to succeed and often outperform their mentors’ impact by balancing their research endeavors between developing their mentors’ secondary topics and establishing themselves in new areas. This suggests a form of academic exaptation, where mentees leverage existing knowledge to explore new research directions. It also underscores mentorship’s critical role in not only guiding mentees but also empowering them to explore and innovate beyond established research domains. The regression analysis further corroborates an inverted-U relationship between mentor-mentee topic dissimilarity and mentees’ scientific impact, accounting for alternative confounding factors. We also observe that overreliance on mentors, marked by strong collaboration, is detrimental to the mentees’ journey toward independence. Therefore, our analysis sheds light on the complex interplay between mentorship guidance, strategic exploration, and mentee independence, which is crucial for nurturing academic leaders capable of contributing novel insights and expanding the boundaries of their disciplines.

While offering valuable insights into academic mentorship, these results have limitations that should be considered. The data sources, the Academic Family Tree and OpenAlex, are extensive but do not comprehensively capture all mentor-mentee relationships across disciplines and cultures, potentially introducing selection bias into the analysis. Despite this, the AFT dataset remains the most extensive mentorship resource available in scientometric research, and previous studies’ reviews of the AFT dataset have not uncovered significant biases that might impact our results[10, 9, 49, 67]. The study’s methodology, using network science techniques and citation analysis, overlooks the works that receive fewer citations yet are significant. This could bias the analysis towards more popular topics. A more inclusive analysis could offer a broader view of the academic landscape. This could be achieved by using text-based topic modeling to include a wider array of mentee contributions[55]. Additionally, we merely use topic-specific impact metrics to quantify the mentees’ career achievements, which may not reflect implicit and qualitative aspects of research performance[68]. These include conceptual frameworks, research design, academic presentations, and the intrinsic abilities of independent and critical thinking, supervision, team collaboration, and communication, which are equally crucial to mentees’ citation impact. Furthermore, external factors such as institutional prestige, collaborator expertise, geographical location, and gender disparities may interact with mentorship to influence a mentee’s academic success, aspects that were not comprehensively analyzed in our study[69, 17, 70, 71, 72, 73]. Lastly, it should be noted that part of the tendency we observe — especially among outperforming mentees — to progressively move away from their mentors’ primary topics may be explained by an increasingly frequent tendency among scientists to switch topics during their careers [13].

Our findings have profound implications for the academic system at all levels. Demonstrating that mentorship is significantly associated with the mentees’ aspirations to explore and excel beyond their mentors’ domains, especially among elite researchers, highlights the need for mentorship practices that prioritize innovation and independent exploration. For the academic system, this emphasizes the importance of fostering diverse and dynamic research agendas that encourage new discoveries and innovations. Concerning individual researchers, our findings highlight a potential roadmap towards above-average academic success, suggesting that mentees should begin by producing their initial work in their mentors’ secondary fields, and then progressively establish themselves as independent researchers either in these secondary fields or even in fields their mentors did not work on. On the one hand, this seems intuitive. On the other hand, such a roadmap is likely associated with both higher rewards and higher risks. Indeed, plenty of evidence demonstrates that switching between [13] and/or exploring new topics [46] is associated with lower impact. In this respect, it should be noted that our study is necessarily limited to scholars with sufficiently long-lived careers, and we do not have any straightforward way to quantify the volume of mentees who leave academia after attempting the aforementioned roadmap. All in all, our findings suggest that developing substantial independence from one’s mentors is a rewarding strategy in the long run conditional on navigating its early setbacks. Notably, this is very much reminiscent of the trajectory observed for many researchers attempting to launch an interdisciplinary career [74].

Several critical considerations arise as we look to the future. Mentees should find a way to balance the utilization of their mentor’s knowledge and social resources with reducing reliance to achieve independence and success. Striking this balance can lead to a more robust and sustainable academic career, contributing to the mentees’ long-term success and the advancement of their fields. Additionally, fostering diversity and innovation in research fields while expanding upon academic traditions is crucial. Digital platforms and social media play a significant role in modernizing the mentorship system, breaking down traditional barriers, and facilitating cross-disciplinary knowledge flow. Encouraging young scholars to explore high-risk but potentially high-reward new research areas without sacrificing depth and quality is essential. In sum, our investigation into the mentor-mentee dynamic of research topics not only underscores the pivotal role of mentorship in shaping academic trajectories but also invites a broader reevaluation of mentorship’s function and practices in nurturing a vibrant, innovative academic prospect.

Methods

Data source.

Our analysis is based on two distinct datasets. The first one is curated from the Academic Family Tree (AFT, Supplementary S1.1.1), an online website (Academictree.org) for collecting mentor-mentee relationships in a crowd-sourced fashion. AFT initially focused on Neuroscience and expanded later to span more than 50 disciplines. The second data set is the OpenAlex (https://openalex.org/, Methods and Supplementary S1.1.2), a bibliographic database containing entities about authors, works (journals, conferences, etc.), affiliations, and citations. One advantage of OpenAlex over other publication databases is that all entities have been disambiguated and associated with identifiers. The name disambiguation accuracy for these authors also has been validated as reliable and unproblematic using extensive and strict procedures in prior works[49, 75] (Supplementary S1.2.3). The AFT and OpenAlex data sets have been connected by matching the same scientists in each data set, and this matching has been validated with extensive and strict procedures[49]. The combined data of AFT and OpenAlex is taken from [49]. In this paper, we conduct our analysis on researchers in Chemistry, Physics, and Neuroscience, amounting to ~0.5 million mentor-mentee pairs, and to ~80k scientists who published ~10 million papers. We motivate our choice for the studied fields in Supplementary Note 1 (Data and preprocessing).

Community detection.

The co-citation network of a pair of mentor and mentee is constructed by linking two papers if they are cited by at least one following work. For simplicity, we do not weigh the links and only consider the topology of the network to conduct community detection. The community structure of the network is detected with the fast unfolding algorithm[53], which is a heuristic method based on modularity optimization. The modularity function considered in this paper is defined as:

\displaystyle Q=\frac{1}{2m}\sum_{i,j}[A_{i,j}-\gamma\frac{k_{i}k_{j}}{2m}]% \delta(c_{i},c_{j}),

(3)

where $A_{ij}$ is an element of the adjacency matrix of the co-citation network, $k_{i}$ is the degree of node $i$ , $m$ is the total number of links in the network, $c_{i}$ is the community to which node $i$ is assigned, the $\delta$ function $\delta(c_{i},c_{j})$ is 1 if $c_{i}=c_{j}$ , and 0 otherwise. The communities are obtained when the function $Q$ is maximized. Note that $\gamma$ is a resolution parameter in $Q$ . We set it as the standard modularity function, i.e., $\gamma$ = 1, in this paper.

Average network distance.

In the co-citation network of papers authored by the mentor and mentee, nodes are connected if they are co-cited by at least one subsequent paper, suggesting a degree of similarity in their content. The longer the path between two nodes, the less similar their research interests are considered to be. Therefore, to quantify the dissimilarity in research interests between mentor and mentee, as shown in Figure 4 g-i, we first identify the shortest paths between any two nodes. We then calculate the average length ( $L$ ) of these shortest paths for paired nodes belonging to either the mentor or the mentee.

\displaystyle L=\frac{1}{|E|\cdot|R|}\sum_{e\in E,\,r\in R}d(e,r)

(4)

where the $E$ and $R$ represent the sets of mentee’s papers and mentor’s papers in their co-citation network, respectively. $d(e,r)$ is the shortest path between node e and node r.

Specifically, if node e and node r are not connected by any path, the length of the shortest path between them is considered to be equal to the length of the longest shortest path in the network, as calculated by the formula 5.

\displaystyle d(e,r)=\max_{i,j\in(E\cup R)}\left(d(i,j)\right)

(5)

Accession codes and data.

The data and code necessary to reproduce the main and supplementary results will be shared in a permanent repository.

References

[1] Sambunjak, D., Straus, S. E. & Marušić, A. Mentoring in academic medicine: a systematic review. \JournalTitleJama 296, 1103–1115 (2006).
[2] Heinisch, D. P. & Buenstorf, G. The next generation (plus one): an analysis of doctoral students’ academic fecundity based on a novel approach to advisor identification. \JournalTitleScientometrics 117, 351–380 (2018).
[3] Pfund, C., Byars-Winston, A., Branchaw, J., Hurtado, S. & Eagan, K. Defining attributes and metrics of effective research mentoring relationships. \JournalTitleAIDS and Behavior 20, 238–248 (2016).
[4] Zimmermann, E. M. et al. Characteristics of faculty at risk of leaving their medical schools: an analysis of the standpoint™ faculty engagement survey. \JournalTitleJournal of healthcare leadership 1–10 (2020).
[5] Yin, H. L. et al. Sustaining the clinical and translational research workforce: training and empowering the next generation of investigators. \JournalTitleAcademic Medicine 90, 861–865 (2015).
[6] Fortunato, S. et al. Science of science. \JournalTitleScience 359, eaao0185 (2018).
[7] Wang, D. & Barabási, A.-L. The science of science (Cambridge University Press, 2021).
[8] Malmgren, R. D., Ottino, J. M. & Amaral, L. A. N. The role of mentorship in protégé performance. \JournalTitleNature 465, 622–626 (2010).
[9] Liénard, J. F., Achakulvisut, T., Acuna, D. E. & David, S. V. Intellectual synthesis in mentorship determines success in academic careers. \JournalTitleNature communications 9, 4840 (2018).
[10] Ma, Y., Mukherjee, S. & Uzzi, B. Mentorship and protégé success in stem fields. \JournalTitleProceedings of the National Academy of Sciences 117, 14077–14083 (2020).
[11] Bentivoglio, M., Vercelli, A. & Filogamo, G. Giuseppe levi: mentor of three nobel laureates. \JournalTitleJournal of the History of the Neurosciences 15, 358–368 (2006).
[12] Foster, J. G., Rzhetsky, A. & Evans, J. A. Tradition and innovation in scientists’ research strategies. \JournalTitleAmerican sociological review 80, 875–908 (2015).
[13] Zeng, A. et al. Increasing trend of scientists to switch between topics. \JournalTitleNature communications 10, 3439 (2019).
[14] Bredella, M. A., Fessell, D. & Thrall, J. H. Mentorship in academic radiology: why it matters. \JournalTitleInsights into Imaging 10, 107 (2019).
[15] Deanna, R. et al. Community voices: the importance of diverse networks in academic mentoring. \JournalTitleNature Communications 13, 1681 (2022).
[16] Xing, Y., Fan, Y., Sinatra, R. & Zeng, A. Academic mentees succeed in big groups, but thrive in small groups. \JournalTitlearXiv preprint arXiv:2208.05304 (2022).
[17] Liu, M., Bu, Y., Li, D., Ding, Y. & Acuna, D. E. Effects of same-race mentorship preferences on academic performance and survival. \JournalTitlearXiv preprint arXiv:2310.09453 (2023).
[18] Milojević, S., Radicchi, F. & Walsh, J. P. Changing demographics of scientific careers: The rise of the temporary workforce. \JournalTitleProceedings of the National Academy of Sciences 115, 12616–12623 (2018).
[19] Gonzalez, L. R., Brambila, C. N. G. & Veloso, F. Birth of prominent scientists. \JournalTitlePloS one 13 (2018).
[20] Mantai, L. “feeling more academic now”: Doctoral stories of becoming an academic. \JournalTitleThe Australian Educational Researcher 46, 137–153 (2019).
[21] Petersen, A. M., Riccaboni, M., Stanley, H. E. & Pammolli, F. Persistence and uncertainty in the academic career. \JournalTitleProceedings of the National Academy of Sciences 109, 5213–5218 (2012).
[22] Yin, Y., Wang, Y., Evans, J. A. & Wang, D. Quantifying the dynamics of failure across science, startups and security. \JournalTitleNature 575, 190–194 (2019).
[23] Janosov, M., Musciotto, F., Battiston, F. & Iñiguez, G. Elites, communities and the limited benefits of mentorship in electronic music. \JournalTitleScientific reports 10, 3136 (2020).
[24] Zeng, H., Zhao, L. & Ruan, S. How does mentoring affect protégés’ adaptive performance in the workplace: roles of thriving at work and promotion focus. \JournalTitleFrontiers in Psychology 11, 546152 (2020).
[25] Libby, A. M. et al. Grant success for early-career faculty in patient-oriented research: difference-in-differences evaluation of an interdisciplinary mentored research training program. \JournalTitleAcademic Medicine 91, 1666–1675 (2016).
[26] Scandura, T. A. Mentorship and career mobility: An empirical investigation. \JournalTitleJournal of organizational behavior 13, 169–174 (1992).
[27] Hsieh, S.-T. & Hsu, P.-Y. Mentoring effects in the successful adaptation of information systems. \JournalTitleInformation Development 31, 164–175 (2015).
[28] Manthiram, K. & Edwards, K. M. Reflections on the mentor-mentee relationship. \JournalTitleJournal of the Pediatric Infectious Diseases Society 10, 1040–1043 (2021).
[29] Battiston, F. et al. Taking census of physics. \JournalTitleNature Reviews Physics 1, 89–97 (2019).
[30] Myers, K. R. et al. Unequal effects of the covid-19 pandemic on scientists. \JournalTitleNature human behaviour 4, 880–883 (2020).
[31] Andalón, M., de Fontenay, C., Ginther, D. K. & Lim, K. The rise of teamwork and career prospects in academic science. \JournalTitleNature Biotechnology 42, 1314–1319 (2024).
[32] Barres, B. A. Stop blocking postdocs’ paths to success. \JournalTitleNature 548, 517–519 (2017).
[33] Hamilton, J. G. et al. Transitioning to independence and maintaining research careers in a new funding climate: american society of preventive oncology junior members interest group report (2013).
[34] Zhang, S., Wapman, K. H., Larremore, D. B. & Clauset, A. Labor advantages drive the greater productivity of faculty at elite universities. \JournalTitleScience Advances 8, eabq7056 (2022).
[35] Patsali, S., Pezzoni, M. & Visentin, F. Research independence: drivers and impact on phd students’ careers. \JournalTitleStudies in Higher Education 1–24 (2024).
[36] Bulian, L., Čavar, I. & Mance, Z. “it’s dangerous to go alone!“scientific excellence of phd holders and their mentors–network analysis of croatian doctoral students. \JournalTitleInterdisciplinary Description of Complex Systems: INDECS 20, 483–499 (2022).
[37] Rosenfeld, A. & Maksimov, O. Should young computer scientists stop collaborating with their doctoral advisors? \JournalTitleCommunications of the ACM 65, 66–72 (2022).
[38] Liardet, C. L. Navigating the transition into higher degree research: An exploration of candidates’ experiences. \JournalTitleThe Australian Educational Researcher 1–18 (2023).
[39] Venturini, S., Sikdar, S., Rinaldi, F., Tudisco, F. & Fortunato, S. Collaboration and topic switches in science. \JournalTitleScientific Reports 14, 1258 (2024).
[40] Liu, L., Dehmamy, N., Chown, J., Giles, C. L. & Wang, D. Understanding the onset of hot streaks across artistic, cultural, and scientific careers. \JournalTitleNature communications 12, 5392 (2021).
[41] Sun, Y., Caccioli, F., Li, X. & Livan, G. The academic great gatsby curve. \JournalTitleJournal of the Royal Society Interface 21, 20240173 (2024).
[42] Sugimoto, C. R., Ni, C., Russell, T. G. & Bychowski, B. Academic genealogy as an indicator of interdisciplinarity: An examination of dissertation networks in library and information science. \JournalTitleJournal of the American Society for Information Science and Technology 62, 1808–1828 (2011).
[43] Wuestman, M., Frenken, K. & Wanzenböck, I. A genealogical approach to academic success. \JournalTitlePlos one 15, e0243913 (2020).
[44] Wu, L., Wang, D. & Evans, J. A. Large teams develop and small teams disrupt science and technology. \JournalTitleNature 566, 378–382 (2019).
[45] Hofstra, B., McFarland, D. A., Smith, S. & Jurgens, D. Diversifying the professoriate. \JournalTitleSocius 8, 23780231221085118 (2022).
[46] Yegros-Yegros, A., Rafols, I. & D’este, P. Does interdisciplinary research lead to higher citation impact? the different effect of proximal and distal interdisciplinarity. \JournalTitlePloS one 10, e0135095 (2015).
[47] Wikipedia. Exaptation (2024). Accessed: 2024-Aug-12.
[48] Gould, S. J. & Vrba, E. S. Exaptation—a missing term in the science of form. \JournalTitlePaleobiology 8, 4–15 (1982).
[49] Ke, Q., Liang, L., Ding, Y., David, S. V. & Acuna, D. E. A dataset of mentorship in bioscience with semantic and demographic estimations. \JournalTitleScientific data 9, 467 (2022).
[50] Shen, H.-W. & Barabási, A.-L. Collective credit allocation in science. \JournalTitleProceedings of the National Academy of Sciences 111, 12325–12330 (2014).
[51] Xing, Y., Wang, F., Zeng, A. & Ying, F. Solving the cold-start problem in scientific credit allocation. \JournalTitleJournal of Informetrics 15, 101157 (2021).
[52] Ahn, Y.-Y., Bagrow, J. P. & Lehmann, S. Link communities reveal multiscale complexity in networks. \JournalTitlenature 466, 761–764 (2010).
[53] Blondel, V. D., Guillaume, J.-L., Lambiotte, R. & Lefebvre, E. Fast unfolding of communities in large networks. \JournalTitleJournal of statistical mechanics: theory and experiment 2008, P10008 (2008).
[54] Small, H. Co-citation in the scientific literature: A new measure of the relationship between two documents. \JournalTitleJournal of the American Society for information Science 24, 265–269 (1973).
[55] Ding, Y., Chowdhury, G. G. & Foo, S. Bibliometric cartography of information retrieval research by using co-word analysis. \JournalTitleInformation processing & management 37, 817–842 (2001).
[56] Tumminello, M., Micciche, S., Lillo, F., Piilo, J. & Mantegna, R. N. Statistically validated networks in bipartite complex systems. \JournalTitlePloS one 6, e17994 (2011).
[57] Marcaccioli, R. & Livan, G. A pólya urn approach to information filtering in complex networks. \JournalTitleNature communications 10, 745 (2019).
[58] Börner, K., Sanyal, S., Vespignani, A. et al. Network science. \JournalTitleAnnu. rev. inf. sci. technol. 41, 537–607 (2007).
[59] Radicchi, F., Castellano, C., Cecconi, F., Loreto, V. & Parisi, D. Defining and identifying communities in networks. \JournalTitleProceedings of the national academy of sciences 101, 2658–2663 (2004).
[60] Börner, K., Chen, C. & Boyack, K. W. Visualizing knowledge domains. \JournalTitleAnnual review of information science and technology 37, 179–255 (2003).
[61] Ke, Q., Gates, A. J. & Barabási, A.-L. A network-based normalized impact measure reveals successful periods of scientific discovery across disciplines. \JournalTitleProceedings of the National Academy of Sciences 120, e2309378120 (2023).
[62] Waltman, L. & van Eck, N. J. Field-normalized citation impact indicators and the choice of an appropriate counting method. \JournalTitleJournal of Informetrics 9, 872–894 (2015).
[63] Yin, Y. & Wang, D. The time dimension of science: Connecting the past to the future. \JournalTitleJournal of Informetrics 11, 608–621 (2017).
[64] Li, W., Aste, T., Caccioli, F. & Livan, G. Early coauthorship with top scientists predicts success in academic careers. \JournalTitleNature communications 10, 5170 (2019).
[65] Xing, Y., Zeng, A., Fan, Y. & Di, Z. The strong nonlinear effect in academic dropout. \JournalTitleScientometrics 120, 793–805 (2019).
[66] Sekara, V. et al. The chaperone effect in scientific publishing. \JournalTitleProceedings of the National Academy of Sciences 115, 12603–12607 (2018).
[67] Schwartz, L. P., Liénard, J. & David, S. V. Impact of gender on the formation and outcome of mentoring relationships in academic research. \JournalTitlearXiv preprint arXiv:2104.07780 (2021).
[68] Sugimoto, C. R. & Larivière, V. Measuring research: What everyone needs to know (Oxford University Press, 2018).
[69] AlShebli, B. K., Rahwan, T. & Woon, W. L. The preeminence of ethnic diversity in scientific collaboration. \JournalTitleNature communications 9, 5163 (2018).
[70] Spoon, K. et al. Gender and retention patterns among us faculty. \JournalTitleScience advances 9, eadi2205 (2023).
[71] Wapman, K. H., Zhang, S., Clauset, A. & Larremore, D. B. Quantifying hierarchy and dynamics in us faculty hiring and retention. \JournalTitleNature 610, 120–127 (2022).
[72] Larivière, V., Ni, C., Gingras, Y., Cronin, B. & Sugimoto, C. R. Bibliometrics: Global gender disparities in science. \JournalTitleNature 504, 211–213 (2013).
[73] Vásárhelyi, O., Zakhlebin, I., Milojević, S. & Horvát, E.-Á. Gender inequities in the online dissemination of scholars’ work. \JournalTitleProceedings of the National Academy of Sciences 118, e2102945118 (2021).
[74] Sun, Y., Livan, G., Ma, A. & Latora, V. Interdisciplinary researchers attain better long-term funding performance. \JournalTitleCommunications Physics 4, 263 (2021).
[75] Lin, Y., Frey, C. B. & Wu, L. Remote collaboration fuses fewer breakthrough ideas. \JournalTitleNature 623, 987–991 (2023).

Acknowledgements

We thank An Zeng and Chongxin Zhong for their helpful conversations. This work is supported by the National Natural Science Foundation of China under Grants (Nos. 62006109 and 12031005) and the Stable Support Plan Program of Shenzhen Natural Science Fund (No. 20220814165010001).

Author contributions statement

Y.X. and Y.M. conceived the study. All authors contributed to the design of the study. Y.X. curated the datasets and developed the algorithm. Y.X., T.P., and X.L. performed the analysis. Y.X., Y.S., and Y.M. contributed to the interpretation of the results. G.L. and Y.M. supervised the study. Y.X., Y.S., G.L., and Y.M. wrote the manuscript.

Competing interests

The authors declare no competing interests.

Inclusion and ethics

All authors have agreed to all manuscript contents, the author list and its order, and the author contribution statements.