10-fold Cross-Validation Area-Under-the-Curve scores (mean ± st.dev.) of protein representation methods in the inference tasks, under varying experimental setup settings. Hist-8000 matches SoT in six out of seven tasks (tasks 1, 3-8 considered here), after thorough comparisons of the representation methods under different classifier models and their hyper-parameters. Hist-8000 consists of the conceptually simpler BoW approach [13], in contrast to SoT which requires SSL pre-training (word2vec) on a large protein sequence dataset of more than 500k proteins [17]. Best-performing representations in bold, with the classifier used for the top representation method provided in separate column. See section ‘Protein inference problems’ for task data

10-fold Cross-Validation Area-Under-the-Curve scores (mean ± st.dev.) of protein representation methods in the inference tasks, under varying experimental setup settings. Hist-8000 matches SoT in six out of seven tasks (tasks 1, 3-8 considered here), after thorough comparisons of the representation methods under different classifier models and their hyper-parameters. Hist-8000 consists of the conceptually simpler BoW approach [13], in contrast to SoT which requires SSL pre-training (word2vec) on a large protein sequence dataset of more than 500k proteins [17]. Best-performing representations in bold, with the classifier used for the top representation method provided in separate column. See section ‘Protein inference problems’ for task data