Mastering scientific communication (SC) skills is essential for successful scientists, and can be a challenge for junior scientists from any background, but may be especially difficult for those who are educationally under-prepared or who have had limited exposure to standard academic English. The burden of teaching such skills typically falls to mentors, who may find this task labor-intensive and frustrating. Non-native English-speaking mentors may have particular difficulty, but even mentors with excellent language skills may not know how to teach effectively, and addressing these needs can further burden mentoring relationships already challenged by gender, age, or cultural gaps. Thus, the gap to address is the identification and modification of both trainee and mentor factors that influence trainee self-efficacy in SC skills development, particularly if such factors play a role n persistence of underrepresented groups in research. Our aims are to identify causal relationships between trainee and mentor factors in an adapted social cognitive model of SC skill development and to identify key moderating factors that change these relationships over time. We hypothesize that trainee efficacy beliefs in SC will predict their outcome expectations and interests in mastering SC skills and that mentor factors (e.g., high skill level in SC coaching availability, high interest), will be positively associated with trainee SC self-efficacy, goals, ad actions. We also hypothesize that the relationship between such mentor factors, trainee actions, and trainee long-term career goals will be fully mediated by trainee SC performance outcomes, and that mentors who are highly skilled and interested in SC mentoring will have trainees whose SC self-efficacy and performance outcomes will increase or remain high. Recruiting a cohort of 400 doctoral and postdoctoral trainees and their 400 faculty mentors from MD Anderson Cancer Center and through training programs elsewhere enriched for trainees of underrepresented groups, we will collect data 3 times over 30 months. Trainee-mentor dyads will be recruited using a snowball approach until desired numbers are achieved in categories by trainee factors (e.g., gender, race/ethnicity). Data will be collected through a secure website with IDs that match trainee-mentor dyads without personal identifiers, although matching is not required to accomplish study aims. Incentives will be provided for incremental completion of surveys, and to encourage mentor participation, additional gift-cards will be given to trainees when mentors complete surveys. Analytic approaches include structural equation, hierarchical, and longitudinal regression modeling. The multi-disciplinary team includes individuals trained in English, linguistics, epidemiology, social psychology, education, and communication. If successful, we will have a validated conceptual model describing trainee SC skill acquisition and its impact on persistence in research careers, and the mediating influence of mentors measured over time. This model will identify risk points for us to develop educational interventions that increase persistence of trainees in pursuing research careers.