Rare Cell-Population Detection in scRNA-Seq

How reliably can unsupervised methods find a rare cell population in single-cell RNA-seq, and what happens as the population gets rarer? This project spikes cells of known identity (FACS-sorted blood cells, SNP-demultiplexed lung cancer cell lines) into a background at 0.1 % to 5 % and compares eight detectors in 162 controlled experiments: isolation forest, local outlier factor (LOF) and one-class SVM on the PCA latent space, mean kNN distance, Leiden small clusters, GiniClust- and FiRE-style rare-cell scores and a cluster-smoothed isolation forest. The detectors are then checked on naturally rare populations (PBMC 3k, Jurkat:293T), and a web application scores single cells against a PBMC 3k reference. A complete stand-alone MATLAB version (anomaly detection with iforest, lof and ocsvm) is included.

Category: Bioinformatics - Transcriptomics and Gene-Expression Analysis (Machine Learning, Isolation Forest, Clustering, Anomaly Detection, Single-Cell RNA-Seq, MATLAB, Python)

Transcriptomics & Gene-Expression Analysis Project B14 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.