Cell Type Classification

This task evaluates how well model embeddings capture information relevant to cell identity. This is achieved by a forward pass of the data through each model to retrieve embeddings, and then using the embeddings to train different classifiers to predict cell type, in this case we are using Logistic Regression, KNN, and RandomForest. To ensure a reliable evaluation, a 5-fold cross-validation strategy is employed. For each split, the classifier's predictions on the held-out data, along with the true cell type labels, are used to compute a range of classification metrics. The final benchmark output for each metric is the average across all cross-validation folds among the 3 classifiers and over 5 random seeds.