Outcome labeling ambiguity and subjectivity are ubiquitous in real-world datasets. While practitioners commonly combine ambiguous outcome labels for all data points (instances) in an ad hoc way to improve the accuracy of multi-class classification, there lacks a principled approach to guide the label combination for all data points by any optimality criterion. To address this problem, we propose the information-theoretic classification accuracy (ITCA), a criterion that balances the trade-off between prediction accuracy (how well do predicted labels agree with actual labels) and classification resolution (how many labels are predictable), to guide practitioners on how to combine ambiguous outcome labels. To find the optimal label combination indicated by ITCA, we propose two search strategies: greedy search and breadth-first search. Notably, ITCA and the two search strategies are adaptive to all machine-learning classification algorithms. Coupled with a classification algorithm and a search strategy, ITCA has two uses: improving prediction accuracy and identifying ambiguous labels. We first verify that ITCA achieves high accuracy with both search strategies in finding the correct label combinations on synthetic and real data. Then we demonstrate the effectiveness of ITCA in diverse applications, including medical prognosis, cancer survival prediction, user demographics prediction, and cell type classification. We also provide theoretical insights into ITCA by studying the oracle and the linear discriminant analysis classification algorithms. Python package itca (available at https://github.com/JSB-UCLA/ITCA) implements ITCA and the search strategies.

Despite the availability of numerous statistical and machine learning tools for joint feature modeling, many scientists investigate features marginally, i.e., one feature at a time. This is partly due to training and convention but also roots in scientists' strong interests in simple visualization and interpretability. As such, marginal feature ranking for some predictive tasks, e.g., prediction of cancer driver genes, is widely practiced in the process of scientific discoveries. In this work, we focus on marginal ranking for binary classification, one of the most common predictive tasks. We argue that the most widely used marginal ranking criteria, including the Pearson correlation, the two-sample t test, and two-sample Wilcoxon rank-sum test, do not fully take feature distributions and prediction objectives into account. To address this gap in practice, we propose two ranking criteria corresponding to two prediction objectives: the classical criterion (CC) and the Neyman-Pearson criterion (NPC), both of which use model-free nonparametric implementations to accommodate diverse feature distributions. Theoretically, we show that under regularity conditions, both criteria achieve sample-level ranking that is consistent with their population-level counterpart with high probability. Moreover, NPC is robust to sampling bias when the two class proportions in a sample deviate from those in the population. This property endows NPC good potential in biomedical research where sampling biases are ubiquitous. We demonstrate the use and relative advantages of CC and NPC in simulation and real data studies. Our model-free objective-based ranking idea is extendable to ranking feature subsets and generalizable to other prediction tasks and learning objectives.

Data-driven discovery of cancer driver genes, including tumor suppressor genes (TSGs) and oncogenes (OGs), is imperative for cancer prevention, diagnosis, and treatment. Although epigenetic alterations are important for tumor initiation and progression, most known driver genes were identified based on genetic alterations alone. Here, we developed an algorithm, DORGE (Discovery of Oncogenes and tumor suppressoR genes using Genetic and Epigenetic features), to identify TSGs and OGs by integrating comprehensive genetic and epigenetic data. DORGE identified histone modifications as strong predictors for TSGs, and it found missense mutations, super enhancers, and methylation differences as strong predictors for OGs. We extensively validated DORGE-predicted cancer driver genes using independent functional genomics data. We also found that DORGE-predicted dual-functional genes (both TSGs and OGs) are enriched at hubs in protein-protein interaction and drug-gene networks. Overall, our study has deepened the understanding of epigenetic mechanisms in tumorigenesis and revealed previously undetected cancer driver genes.

Survivors of traumatic brain injury (TBI) have an unpredictable clinical course. This unpredictability makes clinical resource allocation for clinicians and anticipatory guidance for patients difficult. Historically, experienced clinicians and traditional statistical models have insufficiently considered all available clinical information to predict functional outcomes for a TBI patient. Here, we harness artificial intelligence and apply machine learning and statistical models to predict the Functional Independence Measure (FIM) scores after rehabilitation for traumatic brain injury (TBI) patients. Tree-based algorithmic analysis of 629 TBI patients admitted to a large acute rehabilitation facility showed statistically significant improvement in motor and cognitive FIM scores at discharge.

The mammalian DNA methylome is formed by two antagonizing processes, methylation by DNA methyltransferases (DNMT) and demethylation by ten-eleven translocation (TET) dioxygenases. Although the dynamics of either methylation or demethylation have been intensively studied in the past decade, the direct effects of their interaction on gene expression remain elusive. Here, we quantify the concurrence of DNA methylation and demethylation by the percentage of unmethylated CpGs within a partially methylated read from bisulfite sequencing. After verifying 'methylation concurrence' by its strong association with the co-localization of DNMT and TET enzymes, we observe that methylation concurrence is strongly correlated with gene expression. Notably, elevated methylation concurrence in tumors is associated with the repression of 40~60% of tumor suppressor genes, which cannot be explained by promoter hypermethylation alone. Furthermore, methylation concurrence can be used to stratify large undermethylated regions with negligible differences in average methylation into two subgroups with distinct chromatin accessibility and gene regulation patterns. Together, methylation concurrence represents a unique methylation metric important for transcription regulation and is distinct from conventional metrics, such as average methylation and methylation variation.

High-throughput biological data analysis commonly involves identifying features such as genes, genomic regions, and proteins, whose values differ between two conditions, from numerous features measured simultaneously. The most widely used criterion to ensure the analysis reliability is the false discovery rate (FDR), which is primarily controlled based on p-values. However, obtaining valid p-values relies on either reasonable assumptions of data distribution or large numbers of replicates under both conditions. Clipper is a general statistical framework for FDR control without relying on p-values or specific data distributions. Clipper outperforms existing methods for a broad range of applications in high-throughput data analysis.