Rankings and leaderboards are often used in crowdsourcing contests and online communities to motivate individual contributions but feedback based on social comparison can also have negative effects. Here, we study the unequal effects of such feedback on individual effort and performance for individuals of different ability. We hypothesize that the effects of social comparison differ for top performers and bottom performers in a way that the inequality between the two increases. We use a quasi-experimental design to test our predictions with data from Topcoder, a large online crowdsourcing platform that publishes computer programming contests. We find that in contests where the submitted code is evaluated against others' submissions, rather than using an absolute scale, top performers increase their effort while bottom performers decrease it. As a result, relative scoring leads to better outcomes for those at the top but lower engagement for bottom performers. Our findings expose an important but overlooked drawback from using gamified competitions, rankings, and relative evaluations, with potential implications for crowdsourcing markets, online learning environments, online communities, and organizations in general.

MOTIVATION: Do machine learning methods improve standard deconvolution techniques for gene expression data? This article uses a unique new dataset combined with an open innovation competition to evaluate a wide range of approaches developed by 294 competitors from 20 countries. The competition's objective was to address a deconvolution problem critical to analyzing genetic perturbations from the Connectivity Map. The issue consists of separating gene expression of individual genes from raw measurements obtained from gene pairs. We evaluated the outcomes using ground-truth data (direct measurements for single genes) obtained from the same samples. RESULTS: We find that the top-ranked algorithm, based on random forest regression, beat the other methods in accuracy and reproducibility; more traditional gaussian-mixture methods performed well and tended to be faster, and the best deep learning approach yielded outcomes slightly inferior to the above methods. We anticipate researchers in the field will find the dataset and algorithms developed in this study to be a powerful research tool for benchmarking their deconvolution methods and a resource useful for multiple applications. AVAILABILITY AND IMPLEMENTATION: The data is freely available at clue.io/data (section Contests) and the software is on GitHub at https://github.com/cmap/gene_deconvolution_challenge. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Cytotoxic T lymphocytes (CTL) are a major factor in the control of HIV replication. CTL arise in acute infection, causing escape mutations to spread rapidly through the population of infected cells. As a result, the virus develops partial resistance to the immune response. The factors controlling the order of mutating epitope sites are currently unknown and would provide a valuable tool for predicting conserved epitopes. In this work, we adapt a well-established mathematical model of HIV evolution under dynamical selection pressure from multiple CTL clones to include partial impairment of CTL recognition, [Formula: see text], as well as cost to viral replication, [Formula: see text]. The process of escape is described in terms of the cost-benefit tradeoff of escape mutations and predicts a trajectory in the cost-benefit plane connecting sequentially escaped sites, which moves from high recognition loss/low fitness cost to low recognition loss/high fitness cost and has a larger slope for early escapes than for late escapes. The slope of the trajectory offers an interpretation of positive correlation between fitness costs and HLA binding impairment to HLA-A molecules and a protective subset of HLA-B molecules that was observed for clinically relevant escape mutations in the Pol gene. We estimate the value of [Formula: see text] from published experimental studies to be in the range (0.01-0.86) and show that the assumption of complete recognition loss ([Formula: see text]) leads to an overestimate of mutation cost. Our analysis offers a consistent interpretation of the commonly observed pattern of escape, in which several escape mutations are observed transiently in an epitope. This non-nested pattern is a combined effect of temporal changes in selection pressure and partial recognition loss. We conclude that partial recognition loss is as important as fitness loss for predicting the order of escapes and, ultimately, for predicting conserved epitopes that can be targeted by vaccines.