Fig. 4: FusOn-pLM prediction of IDR properties and regions.
From: FusOn-pLM: a fusion oncoprotein-specific language model via adjusted rate masking
A FusOn-pLM-IDR models predict asphericity (A), end-to-end radius (Re), radius of gyration (Rg), and polymer scaling exponent (PS) by feeding FusOn-pLM embeddings through an MLP classification head. B FusOn-pLM-IDR predictions vs. true values. The coefficient of determination (R2) between predictions and labels was calculated for each model to assess goodness of fit. C FusOn-pLM-Diso utilizes a Transformer architecture to predict per-residue disorder labels from FusOn-pLM embeddings. D Disorder predictor performance in CAID2 competition when trained on FusOn-pLM vs. ESM-2-650M embeddings34. E FusOn-pLM-Diso performance on test set fusion oncoproteins, based on AlphaFold-pLDDT-derived disorder labels. Data are presented as first and third quartile +/− 1.5*IQR (interquartile range). Median line is indicated in black and circles represent outliers (left). The coefficient of determination (R2) between predictions and labels was calculated for each model to assess goodness of fit (right). F Visualization of FusOn-pLM embedding predictions of disorder propensity on AlphaFold2-predicted structure. Disorder probabilities are shaded according to the legend for interpolation. Source data for this figure are provided in the Source Data file.
