Coarse composition suffices: tabular in-context learning for multi-activity antimicrobial peptide profiling
Antimicrobial peptides (AMPs) often act against multiple pathogen classes, making multi-label activity prediction a more realistic screening target than binary antimicrobial classification. The ESCAPE benchmark formalizes this setting, but leading approaches typically rely on multimodal, structure-conditioned deep models that are costly to train and tune. We show that a simple, sequence-only…
A new study demonstrates that a simple sequence-only approach can achieve comparable or superior results to more complex deep learning methods for predicting multi-activity antimicrobial peptides (AMPs). By combining 330 interpretable sequence descriptors with TabPFN, a tabular foundation model, researchers achieved an mAP-5 score of 77.8% on the ESCAPE benchmark, surpassing the previous best of 72.1%.
This label-powerset TabPFN model performs in-context prediction in a single forward pass without the need for gradient-based training or hyperparameter search. The improvements observed in the study persist under the prior state-of-the-art single-fold training protocol, indicating they are not simply a result of larger training datasets.
Notably, predicted structure is not necessary for making predictions, and performance is not driven by any single descriptor family. Instead, predicted structure is not necessary for making predictions, and performance is not driven by any single descriptor family. Ablations show that ten global physicochemical scalars recover 91% of the performance achieved with the full set of features.
Additionally, explicitly modeling label dependence allows for targeted improvements in the assessment of scarce activities and supports prioritizing which activities to assay next from partial positive evidence.
Written by urgent.news from bioRxiv's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
