Fig. 3: MTX-based coexpression contributes substantially to FUGAsseM-full protein function predictions.
From: Predicting functions of uncharacterized gene products from microbial communities
a–c, Distribution of RF importance scores from the FUGAsseM-full model’s second (data integration) layer. Only GO terms with sufficient performance (resulting in predictions with confidence probability ≥ 0.75) are included for BP (a; n = 14,249 total term–species pairs for prediction), MF (b; n = 18,553 total term–species pairs for prediction) and CC (c; n = 1,623 total term–species pairs for prediction). The full list is provided in Supplementary Table 11. d–f, Distribution of importance scores for successful FUGAsseM models (those that assigned GO annotations to proteins with a prediction probability ≥ 0.75) showing newly predicted annotations that were not used for training, based on accumulated experimental evidence over time (n = 65 total term–species pairs for BP (d), 34 total term–species pairs for MF (e) and 11 total term–species pairs for CC (f)). The full list is provided in Supplementary Table 12. Box plots display the median (line at the 50th percentile), IQR (box spanning the 25th to 75th percentiles), whiskers (extending to 1.5 × the IQR) and mean values (dark points).
