Improving Classification of Cell Types in Acute Myeloid Leukemia with Self-guided Masking Technique

Acute myeloid leukemia (AML) is a rare but important disease. Because it has many different features and behaviors, classifying its cell types with traditional methods is both difficult and costly. Transformer-based foundation models (FMs) are useful for analyzing biological data, and they usually use random masking during training. But normal uniform random masking selects genes without checking how important they are. To solve this, we propose a self-guided masking method. This method learns which gene positions are most useful to mask at each training step. We show that our approach improves FM training and performs better than uniform masking in cell-type classification for AML.