Rank Candidate Research Articles

Remembering past and predicting future: a hybrid recurrent neural network based recommender system.

Traditional recommender systems (RS) assume users’ taste to be static (taste remains same over time) and reactive (a change in taste cannot be predicted and is observed only after it occurs). Further, traditional RS restricts the recommendation process to candidate items generation. This work aims to explore two phases of RS, i.e., Candidate Generation as well as Candidate Ranking. We propose a RS from a multi-objective (short-term prediction, long-term prediction, diversity, and popularity bias) perspective which was previously overlooked. The sequential and non-sequential behavior of users is exploited to predict future behavioral trajectories with the consideration of short-term and long-term prediction using recurrent neural networks and nearest neighbors approach. Further, a novel candidate ranking method is introduced to prevent users from being entangled in recommended items. On multiple datasets, largest being MovieLens (ML) 1M, our model shows excellent results achieving a hit rate and short-term prediction success of 58% and 71% respectively on ML 1M. Further, it implicitly handles two important parameters, i.e., diversity and item popularity with a success rate of 59.22% and 34.28% respectively.

Machine Learning-Assisted Identification and Quantification of Hydroxylated Metabolites of Polychlorinated Biphenyls in Animal Samples.

Laboratory studies of the disposition and toxicity of hydroxylated polychlorinated biphenyl (OH-PCB) metabolites are challenging because authentic analytical standards for most unknown OH-PCBs are not available. To assist with the characterization of these OH-PCBs (as methylated derivatives), we developed machine learning-based models with multiple linear regression (MLR) or random forest regression (RFR) to predict the relative retention times (RRT) and MS/MS responses of methoxylated (MeO-)PCBs on a gas chromatograph-tandem mass spectrometry system. The final MLR model estimated the retention times of MeO-PCBs with a mean absolute error of 0.55 min (n = 121). The similarity coefficients cos θ between the predicted (by RFR model) and experimental MS/MS data of MeO-PCBs were >0.95 for 92% of observations (n = 96). The levels of MeO-PCBs quantified with the predicted MS/MS response factors approximated the experimental values within a 2-fold difference for 85% of observations and 3-fold differences for all observations (n = 89). Subsequently, these model predictions were used to assist with the identification of OH-PCB 95 or OH-PCB 28 metabolites in mouse feces or liver by suggesting candidate ranking information for identifying the metabolite isomers. Thus, predicted retention and MS/MS response data can assist in identifying unknown OH-PCBs.

A follow-up on intermediate-mass black hole candidates in the second LIGO–Virgo observing run with the Bayes Coherence Ratio

ABSTRACT The detection of an intermediate-mass black hole population (102–106 M⊙) will provide clues to their formation environments (e.g. discs of active galactic nuclei, globular clusters) and illuminate a potential pathway to produce supermassive black holes. Ground-based gravitational-wave detectors are sensitive to mergers that can form intermediate-mass black holes weighing up to ∼450 M⊙. However, ground-based detector data contain numerous incoherent short duration noise transients that can mimic the gravitational-wave signals from merging intermediate-mass black holes, limiting the sensitivity of searches. Here, we follow-up on binary black hole merger candidates using a ranking statistic that measures the coherence or incoherence of triggers in multiple-detector data. We use this statistic to rank candidate events, initially identified by all-sky search pipelines, with lab-frame total masses ≳ 55 M⊙ using data from LIGO’s second observing run. Our analysis does not yield evidence for new intermediate-mass black holes. However, we find support for eight stellar-mass binary black holes not reported in the first LIGO–Virgo gravitational wave transient catalogue GWTC-1, seven of which have been previously reported by other catalogues.

Optimal COVID-19 therapeutic candidate discovery using the CANDO platform.

The worldwide outbreak of SARS-CoV-2 in early 2020 caused numerous deaths and unprecedented measures to control its spread. We employed our Computational Analysis of Novel Drug Opportunities (CANDO) multiscale therapeutic discovery, repurposing, and design platform to identify small molecule inhibitors of the virus to treat its resulting indication, COVID-19. Initially, few experimental studies existed on SARS-CoV-2, so we optimized our drug candidate prediction pipelines using results from two independent high-throughput screens against prevalent human coronaviruses. Ranked lists of candidate drugs were generated using our open source cando.py software based on viral protein inhibition and proteomic interaction similarity. For the former viral protein inhibition pipeline, we computed interaction scores between all compounds in the corresponding candidate library and eighteen SARS-CoV proteins using an interaction scoring protocol with extensive parameter optimization which was then applied to the SARS-CoV-2 proteome for prediction. For the latter similarity based pipeline, we computed interaction scores between all compounds and human protein structures in our libraries then used a consensus scoring approach to identify candidates with highly similar proteomic interaction signatures to multiple known anti-coronavirus actives. We published our ranked candidate lists at the very beginning of the COVID-19 pandemic. Since then, 51 of our 276 predictions have demonstrated anti-SARS-CoV-2 activity in published clinical and experimental studies. These results illustrate the ability of our platform to rapidly respond to emergent pathogens and provide greater evidence that treating compounds in a multitarget context more accurately describes their behavior in biological systems.

Prediction and evaluation of combination pharmacotherapy using natural language processing, machine learning and patient electronic health records

Combination pharmacotherapy targets key disease pathways in a synergistic or additive manner and has high potential in treating complex diseases. Computational methods have been developed to identifying combination pharmacotherapy by analyzing large amounts of biomedical data. Existing computational approaches are often underpowered due to their reliance on our limited understanding of disease mechanisms. On the other hand, observable phenotypic inter-relationships among thousands of diseases often reflect their underlying shared genetic and molecular underpinnings, therefore can offer unique opportunities to design computational models to discover novel combinational therapies by automatically transferring knowledge among phenotypically related diseases. We developed a novel phenome-driven drug discovery system, named TuSDC, which leverages knowledge of existing drug combinations, disease comorbidities, and disease treatments of thousands of disease and drug entities extracted from over 31.5 million biomedical research articles using natural language processing techniques. TuSDC predicts combination pharmacotherapy by extracting representations of diseases and drugs using tensor factorization approaches. In external validation, TuSDC achieved an average precision of 0.77 for top ranked candidates, outperforming a state of art mechanism-based method for discovering drug combinations in treating hypertension. We evaluated top ranked anti-hypertension drug combinations using electronic health records of 84.7 million unique patients and showed that a novel drug combination hydrochlorothiazide-digoxin was associated with significantly lower hazards of subsequent hypertension as compared to the monotherapy hydrochlorothiazide alone (HR: 0.769, 95% CI [0.732, 0.807]) and digoxin alone (0.857, 95% CI [0.785, 0.936]). Data-driven informatics analyses reveal that the renin-angiotensin system is involved in the synergistical interactions of hydrochlorothiazide and digoxin on regulating hypertension. The prediction model’s code with PyTorch version 1.5 is available at http://nlp.case.edu/public/data/TuSDC/.

The effect of virtual interviews on the musculoskeletal radiology fellowship match.

ObjectiveResulting from the COVID-19 pandemic, a virtual interview format was adopted in lieu of in-person interviews for fellowship recruitment. The purpose of this study is to analyze the potential effect that virtual interviews have on musculoskeletal radiology fellowship match results and collect opinions of the process.Materials and methodsAn anonymous survey was sent to 87 listed US-based musculoskeletal radiology fellowship directors, who were asked for their 2019–2020 and 2020–2021 match data: fellowship positions, interviewees, ranked interviewees, and matched ranks. They rated effectiveness of virtual interviews on a ten-point scale (1 poor–10 excellent) and provided an ideal percentage of future interviews to be conducted virtually.ResultsThirty responses were received (34.4% response rate). Primary analysis found significant increase in the mean number of candidates ranked in 2020–2021, 14.6 to 17.5 (p = 0.047). The mean rank of matched applicants increased from 6.02 to 6.43, without significance (p = 0.821). Secondary analysis, which attempted to exclude internally matched programs, found significant increases in 2020–2021 in the mean number of applicants—28.7 to 32.4 (p = 0.017), interviews conducted—17.8 to 21.3 (p = 0.007), and candidates ranked by programs—16.3 to 19.8 (p = 0.015). The mean rank of matched applicants increased from 6.39 to 7.03, without significance (p = 0.713).ConclusionWith results showing an increase in applications, interviews conducted, and ranked applicants while lowering the average rank of matched candidates, musculoskeletal radiology fellowship directors should consider interviewing more applicants than they usually would in the prior in-person recruitment format.Supplementary InformationThe online version contains supplementary material available at 10.1007/s00256-022-04155-w.

Blue Skies With Clouds: Envisioning the Future Ideal State and Identifying Ongoing Tensions in the UME-GME Transition.

The transition from medical school to residency in the United States consumes large amounts of time for students and educators in undergraduate and graduate medical education (UME, GME), and it is costly for both students and institutions. Attempts to improve the residency application and Match processes have been insufficient to counteract the very large number of applications to programs. To address these challenges, the Coalition for Physician Accountability charged the Undergraduate Medical Education to Graduate Medical Education Review Committee (UGRC) with crafting recommendations to improve the system for the UME-GME transition. To guide this work, the UGRC defined and sought stakeholder input on a "blue-skies" ideal state of this transition. The ideal state views the transition as a system to support a continuum of professional development and learning, thus serving learners, educators, and the public, and engendering trust among them. It also supports the well-being of learners and educators, promotes diversity, and minimizes bias. This manuscript uses polarity thinking to analyze 3 persistent key tensions in the system that require ongoing management. First, the formative purpose of assessment for learning and growth is at odds with the use of assessment data for ranking and sorting candidates. Second, the function of residents as learners can conflict with their role as workers contributing service to health care systems. Third, the current residency Match process can position the desire for individual choice-among students and their programs-against the workforce needs of the profession and the public. This Scholarly Perspective presents strategies to balance the upsides and downsides inherent to these tensions. By articulating the ideal state of the UME-GME transition and anticipating tensions, educators and educational organizations can be better positioned to implement UGRC recommendations to improve the transition system.

Gender Distribution of Authors of Evaluation Letters for Promotion at One Medical School.

Women remain underrepresented in key leadership positions and advanced ranks in academic medicine. This study examines the numbers of men and women letter writers for promotion candidates during a 5-year period across departments, tracks, ranks, and candidate gender. A descriptive study characterized the gender of evaluation letter writers for candidates for promotion to associate or full professor at the University of Minnesota Medical School between 2015 and 2020. Letter writer and candidate gender were characterized by self-identified pronouns in the faculty biography or dossier. Letter writer gender was described by candidate department, promotion track, rank, terminal degree, and gender. Among 299 candidates for promotion, 172 (58%) were men and 127 (42%) were women; dossiers included 3,995 evaluation letters. Across all years, men wrote more letters than women (external letters, range, 69% in 2019-2020 to 75% in 2015-2016; internal letters, range, 67% in 2018-2019 to 77% in 2015-2016). Candidates in the family medicine and pediatrics departments had the highest percentages of letters written by women (44% and 40%, respectively). No differences were found in the number of women letter writers by candidate promotion track; however, differences were found by candidate rank (associate professor, 30%; full professor, 23%) and terminal degree (MD/DO, 25%; PhD, 33%; MD-PhD, 20%). Regardless of candidate gender, most evaluation letters were written by men. Women candidates had 15% to 20% more letters authored by women than men candidates (34%-40% vs 18%-23%). The gender pattern of letter writers may reflect implicit biases regarding gender and perceived leadership status, expertise, and success. Adopting policies that promote or require gender diversity among letter writers for promotion candidates may provide an opportunity to encourage faculty to seek diverse networks and recognize the achievements of women faculty.

An Evaluation of Non-Uniform Grade Distribution with the Emergency Medicine Off-Service Standardized Letters of Evaluation.

Introduction: Standardized Letters of Evaluation (SLOEs) are designed to objectively compare medical students to their peers for completed emergency medicine (EM) rotations to be used in the EM residency match. In an attempt to adapt quickly to the lack of availability of in-person EM rotations due to COVID restrictions, “off-service” SLOEs (OSLOEs) were allowed in place of traditional SLOEs. The purpose of this study was to assess the utility of OSLOEs for candidate selection during the 2020-21 application cycle at a single EM residency.Methods: A retrospective cohort review of all OSLOEs submitted during the 2020-21 academic year to an EM residency program was performed. A total of 270 OSLOES were eligible for review. Summary statistics were calculated for the study variables recorded, including global rank, grade, categorical details, and rank.Results: Of the 270 OSLOEs reviewed, 61.9% ranked candidates in the top 10% of their class, with 95% being ranked in the top two categories. Over 90% of students were graded as honors or high pass and over 75% of students were ranked in the top 1/3 for each specific OSLOE category.Conclusion: Our findings reveal questionable utility of the objective measures in the OSLOE as there are signs it may suffer from non-uniform grade distribution, leading to low utility for candidate selection. Our data shows marked over-ranking within the highest 2 categories. EM program directors and faculty should use caution as the OSLOE may not carry the same weight as a traditional SLOE when objectively evaluating prospective students for a match into EM.

Epigenetic profiling linked to multisystem inflammatory syndrome in children (MIS-C): A multicenter, retrospective study.

SummaryBackgroundMost children and adolescents infected with the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) remain asymptomatic or develop a mild coronavirus disease 2019 (COVID-19) that usually does not require medical intervention. However, a small proportion of pediatric patients develop a severe clinical condition, multisystem inflammatory syndrome in children (MIS-C). The involvement of epigenetics in the control of the immune response and viral activity prompted us to carry out an epigenomic study to uncover target loci regulated by DNA methylation that could be altered upon the appearance of MIS-C.MethodsPeripheral blood samples were recruited from 43 confirmed MIS-C patients. 69 non-COVID-19 pediatric samples and 15 COVID-19 pediatric samples without MIS-C were used as controls. The cases in the two groups were mixed and divided into discovery (MIS-C = 29 and non-MIS-C = 56) and validation (MIS-C = 14 and non-MIS-C = 28) cohorts, and balanced for age, gender and ethnic background. We interrogated 850,000 CpG sites of the human genome for DNA methylation variants.FindingsThe DNA methylation content of 33 CpG loci was linked with the presence of MIS-C. Of these sites, 18 (54.5%) were located in described genes. The top candidate gene was the immune T-cell mediator ZEB2; and others highly ranked candidates included the regulator of natural killer cell functional competence SH2D1B; VWA8, which contains a domain of the Von Willebrand factor A involved in the pediatric hemostasis disease; and human leukocyte antigen complex member HLA-DRB1; in addition to pro-inflammatory genes such as CUL2 and AIM2. The identified loci were used to construct a DNA methylation profile (EPIMISC) that was associated with MIS-C in both cohorts. The EPIMISC signature was also overrepresented in Kawasaki disease patients, a childhood pathology with a possible viral trigger, that shares many of the clinical features of MIS-C.InterpretationWe have characterized DNA methylation loci that are associated with MIS-C diagnosis. The identified genes are likely contributors to the characteristic exaggerated host inflammatory response observed in these patients. The described epigenetic signature could also provide new targets for more specific therapies for the disorder.FundingUnstoppable campaign of Josep Carreras Leukaemia Foundation, Fundació La Marató de TV3, Cellex Foundation and CERCA Programme/Generalitat de Catalunya.

Selection of Candidates for Academic Scholarships Using Analytical Hierarchy Process (Ahp) and Simple Additive Weighting (SAW) Methods at National University

The National University administers several scholarships for new students each year. The problem is, the University does not yet have a standard method for determining beneficiaries. They have difficulty in determining prospective scholarship recipients with the same criteria. In fact, sometimes it only relies on people's instincts, which can be subjective. The University should implement a DSS system to overcome this problem. Therefore, in this study, the AHP method is applied and used to weight the criteria and test the level of consistency so that the criteria are in accordance with the type of scholarship. In addition, the SAW method is also used for determining scholarship recipients according to the quota. The results of this study are the priority weights of the importance of each criterion, namely the average value of report cards (0.35), parents' income (0.23), certificates (0.05), affidavits of not working and receiving scholarships (0.15), year of graduation (0.11) and number of certificates (0.10). The consistency ratio value is 0.03741 indicating the weight is consistent. This study also resulted in the best ranking of candidates for academic scholarships, with a result of 0.93. The average value of the application test results using the TAM method is 80.5%.

Ranking reprogramming factors for cell differentiation.

Transcription factor over-expression is a proven method for reprogramming cells to a desired cell type for regenerative medicine and therapeutic discovery. However, a general method for the identification of reprogramming factors to create an arbitrary cell type is an open problem. Here we examine the success rate of methods and data for differentiation by testing the ability of nine computational methods (CellNet, GarNet, EBseq, AME, DREME, HOMER, KMAC, diffTF and DeepAccess) to discover and rank candidate factors for eight target cell types with known reprogramming solutions. We compare methods that use gene expression, biological networks and chromatin accessibility data, and comprehensively test parameter and preprocessing of input data to optimize performance. We find the best factor identification methods can identify an average of 50-60% of reprogramming factors within the top ten candidates, and methods that use chromatin accessibility perform the best. Among the chromatin accessibility methods, complex methods DeepAccess and diffTF have higher correlation with the ranked significance of transcription factor candidates within reprogramming protocols for differentiation. We provide evidence that AME and diffTF are optimal methods for transcription factor recovery that will allow for systematic prioritization of transcription factor candidates to aid in the design of new reprogramming protocols.

KidSpell: Making a difference in spellchecking for children

Children’s ability to spell effectively is a major barrier to using search engines successfully. While search engines make use of spellcheckers to provide spelling corrections to their users, they are designed for more traditional users (i.e., adults) and have proven inadequate for children. The specific target of children for this research are those with early literacy skills (whose are typically ages 6–12). The aim of this work is twofold: first, to address the types of spelling errors children make by researching, developing, and evaluating algorithms to generate and rank candidate English spelling suggestions for children; and second, to improve children’s user experience when using our proposed spellchecker by involving them in the design process through participatory design and evaluating the impact of interactive elements on children’s spellchecking behaviors. The outcomes of our studies and assessments result in a phonetic-based spelling correction model (KidSpell) that can more accurately correction children’s spelling errors than existing state-of-the-art models. Further, we learned that visual and audio cues have a positive impact on children’s ability to find their intended word from a list of spelling suggestions.

Innovations in the prevention and treatment of postpartum hemorrhage: Analysis of a novel medicines development pipeline database.

BackgroundA significant barrier to improving prevention and treatment of postpartum hemorrhage (PPH) is a lack of innovative medicines that meet the needs of women and providers, particularly those in low‐and middle‐income countries (LMICs). The Accelerating Innovation for Mothers (AIM) project established a new database of candidate medicines under development for five pregnancy‐related conditions between 2000 and 2021.ObjectiveTo systematically identify and rank candidates for prevention and treatment of PPH.Search StrategyAdis Insight, Pharmaprojects, WHO ICTRP, PubMed, and grant databases were searched to develop the AIM database.Selection CriteriaAIM database was searched for candidates being evaluated for PPH prevention and treatment, regardless of phase.Data Collection and AnalysisCandidates were ranked as high, medium, or low potential based on prespecified criteria. Analysis was primarily descriptive, describing candidates and development potential.Main ResultsOf the 444 unique candidates, only 39 pertained to PPH. One was high potential (heat‐stable/inhaled oxytocin) and three were medium potential (melatonin, vasopressin and dofetilide via nanoparticle delivery).ConclusionThe pipeline for new PPH medicines is concerningly limited, lacking diversity, and showing little evidence of novel technologies. Without significant investment in early‐phase research, it is unlikely that new products will emerge.

Signaling repurposable drug combinations against COVID-19 by developing the heterogeneous deep herb-graph method.

Coronavirus disease 2019 (COVID-19) has spurred a boom in uncovering repurposable existing drugs. Drug repurposing is a strategy for identifying new uses for approved or investigational drugs that are outside the scope of the original medical indication. Current works of drug repurposing for severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) are mostly limited to only focusing on chemical medicines, analysis of single drug targeting single SARS-CoV-2 protein, one-size-fits-all strategy using the same treatment (same drug) for different infected stages of SARS-CoV-2. To dilute these issues, we initially set the research focusing on herbal medicines. We then proposed a heterogeneous graph embedding method to signaled candidate repurposing herbs for each SARS-CoV-2 protein, and employed the variational graph convolutional network approach to recommend the precision herb combinations as the potential candidate treatments against the specific infected stage. We initially employed the virtual screening method to construct the 'Herb-Compound' and 'Compound-Protein' docking graph based on 480 herbal medicines, 12,735 associated chemical compounds and 24 SARS-CoV-2 proteins. Sequentially, the 'Herb-Compound-Protein' heterogeneous network was constructed by means of the metapath-based embedding approach. We then proposed the heterogeneous-information-network-based graph embedding method to generate the candidate ranking lists of herbs that target structural, nonstructural and accessory SARS-CoV-2 proteins, individually. To obtain precision synthetic effective treatments forvarious COVID-19 infected stages, we employed the variational graph convolutional network method to generate candidate herb combinations as the recommended therapeutic therapies. There were 24 ranking lists, each containing top-10 herbs, targeting 24 SARS-CoV-2 proteins correspondingly, and 20 herb combinations were generated as the candidate-specific treatment to target the four infected stages. The code and supplementary materials are freely available at https://github.com/fanyang-AI/TCM-COVID19.

Anaphoric Pronouns and the Computation of Prominence Profiles

Previous research has investigated anaphoric resolution at the anaphor. Using a self-paced reading study we show that prominence profiles, i.e. the ranking of the referential candidates for anaphoric resolution, are dynamically established as discourse unfolds. We compared four types of context sentences introducing two referents and found that the cost of the computation of the prominence profile depends on the alignment of prominence-lending features, namely ‘left edge’, ‘agent’, ‘subject’. Cost occurs as referents become available. Further downstream, we contrasted two types of pronouns in German, personal pronoun vs. demonstrative pronoun. By the time the pronoun is encountered, profile computation is already complete, as indicated by the lack of interaction between context and pronoun type. An effect of pronoun reveals that resolution is driven by the form-dependent strength with which an interpretation is obtained (demonstrative pronouns being more stable than personal pronouns). The results also indicate that two prominence-lending features – subjecthood and agentivity – compete with each other.

The Effect of Changing Fellowship Interview Format on Candidate Ranking Variabilities: The COVID-19 Experience.

Purpose This study investigated the effect of interview format changes (in-person to virtual, one-to-one to multiple-to-one) necessitated by the COVID-19 travel restrictions on preliminary fellowship candidate ranking variabilities. Design Cross-sectional observational study. Method In 2018 and 2019, the glaucoma fellowship interviews were conducted in-person in a one-to-one format, whereas in 2020, interviews were virtual and in a multiple (interviewers)-to-one (candidate) format. We compared ranking ranges of interviewers within the same virtual room (WSR) and not within the same virtual room (NWSR) to assess the effect of WSR versus NWSR on ranking variabilities. We also compared ranking categories (“accept,” “alternate,” and “pass”) agreements between 2018, 2019, and 2020 to assess the effect of virtual versus in-person interviews on ranking variabilities. Results NWSR and WSR mean rankings differed by 1.33 (95% confidence interval difference 0.61 to 2.04, p = 0.0003), with WSR interviewers having less variability than NWSR pairs. The variability between 2018/2019 (in-person interviews) and 2020 (virtual interviews) showed no differences between in-person and virtual interviews (weighted Kappa statistic 0.086 for 2018, 0.158 for 2019, and 0.101 for 2020; p < 0.05 for all years). The overall least attractive candidate has the lowest variability; the most attractive candidate has the second lowest variability. Conclusion Grouping interviewers WSR during the interview decreased ranking variabilities compared to NWSR, while a change from in-person to virtual interview format did not increase the ranking variabilities. This suggests that the decreased nonverbal interactions in virtual interviews do not decrease interviewers' perceptions as applied to preliminary rankings.

The Extended Tidal Tails of NGC 7089 (M2)

Using photometry and proper motions from Gaia Early Data Release 3, we detect a 45° long trailing stellar debris stream associated with the old, metal-poor globular cluster NGC 7089. With a width on the order of 100 pc, the extended stream appears to be as dynamically cold as the coldest known streams found to date. There is some evidence for an extended leading tail extending between 28° and 37° from the cluster, though the greater distance of this tail, combined with proper motions that are virtually indistinguishable from those of foreground stars, make the detection much less certain. The proper motion profile and the path on the sky of the trailing tail are not well matched using a simple Galactic potential composed purely of a disk, bulge, and spherical halo. However, the addition of a moving, massive (M = 1.88 × 1011 M ⊙) Large Magellanic Cloud brings the model predictions into much better agreement with the observables. We provide tables of the most highly ranked candidate stream stars for follow-up by ongoing and future spectroscopic surveys.

Machine Learning for The Prediction of Ranked Applicants and Matriculants to an Internal Medicine Residency Program

Phenomenon : Residency programs throughout the country each receive hundreds to thousands of applications every year. Holistic review of this many applications is challenging, and to-date, few tools exist to streamline or assist in the process for selecting candidates to interview and rank. Machine learning could assist programs in predicting which applicants are likely to be ranked, and among ranked applicants, which are likely to matriculate. Approach : In the present study, we used the machine learning algorithm Random Forest (RF) to differentiate between ranked and unranked applicants as well as matriculants and ranked non-matriculants to an internal medicine residency program in northern New England over a three-year period. In total, 5,067 ERAS applications were received during the 2016-17, 2017-18, and 2018-19 application cycles. Of these, 4,256 (84.0%) were unranked applicants, 754 (14.9%) were ranked non-matriculants, and 57 (1.12%) were ranked matriculants. Findings : For differentiating between ranked and unranked applicants, the RF algorithm achieved an area under the receiver operating characteristic (AUROC) curve of 0.925 (95% CI: 0.918-0.932) and area under the precision-recall curve (AUPRC) of 0.652 (0.611-0.685), while for differentiating between matriculants and ranked non-matriculants, the AUROC was 0.597 (95% CI: 0.516-0.680) and AUPRC was 0.114 (0.075-0.167). The ranks of matriculated applicants were significantly higher using the algorithmic rank list as compared with the actual rank list for the 2017-18 (median rank: 98 versus 204, p < .001) and 2018-19 cycles (74 versus 192, p = .006), but not the 2016-17 cycle (97 versus 144, p = .37). Insights : The RF algorithm predicted which applicants among the overall applicant pool were ranked with impressive accuracy and identified matriculants among ranked candidates with modest but better-than-random accuracy. This approach could assist residency programs with triaging applicants based on the likelihood of a candidate being ranked and/or matriculating.

Effect of genotyping strategies on the sustained benefit of single-step genomic BLUP over multiple generations

BackgroundSingle-step genomic best linear unbiased prediction (ssGBLUP) allows the inclusion of information from genotyped and ungenotyped individuals in a single analysis. This avoids the need to genotype all candidates with the potential benefit of reducing overall costs. The aim of this study was to assess the effect of genotyping strategies, the proportion of genotyped candidates and the genotyping criterion to rank candidates to be genotyped, when using ssGBLUP evaluation. A simulation study was carried out assuming selection over several discrete generations where a proportion of the candidates were genotyped and evaluation was done using ssGBLUP. The scenarios compared were: (i) three genotyping strategies defined by their protocol for choosing candidates to be genotyped (RANDOM: candidates were chosen at random; TOP: candidates with the best genotyping criterion were genotyped; and EXTREME: candidates with the best and worse criterion were genotyped); (ii) eight proportions of genotyped candidates (p); and (iii) two genotyping criteria to rank candidates to be genotyped (candidates’ own phenotype or estimated breeding values). The criteria of the comparison were the cumulated gain and reliability of the genomic estimated breeding values (GEBV).ResultsThe genotyping strategy with the greatest cumulated gain was TOP followed by RANDOM, with EXTREME behaving as RANDOM at low p and as TOP with high p. However, the reliability of GEBV was higher with RANDOM than with TOP. This disparity between the trend of the gain and the reliability is due to the TOP scheme genotyping the candidates with the greater chances of being selected. The extra gain obtained with TOP increases when the accuracy of the selection criterion to rank candidates to be genotyped increases.ConclusionsThe best strategy to maximise genetic gain when only a proportion of the candidates are to be genotyped is TOP, since it prioritises the genotyping of candidates which are more likely to be selected. However, the strategy with the greatest GEBV reliability does not achieve the largest gain, thus reliability cannot be considered as an absolute and sufficient criterion for determining the scheme which maximises genetic gain.