Unsupervised multi-instance learning for protein structure determination

Unsupervised Machine Learning Organization of the Functional Dark Proteome of Gram-Negative "Superbugs": Six Protein Clusters Amenable for Distinct Scientific Applications

Uncharacterized proteins have been underutilized as targets for the development of novel therapeutics for difficult-to-treat bacterial infections. To facilitate the exploration of these proteins, 2819 predicted, uncharacterized proteins (19.1% of the total) from reference strains of multidrug Acinetobacter baumannii, Klebsiella pneumoniae, and Pseudomonas aeruginosa species were organized using an unsupervised k-means machine learning algorithm. Classification using normalized values for protein length, pI, hydrophobicity, degree of conservation, structural disorder, and %AT of the coding gene rendered six natural clusters. Cluster proteins showed different trends regarding operon membership, expression, presence of unknown function domains, and interactomic relevance. Clusters 2, 4, and 5 were enriched with highly disordered proteins, nonworkable membrane proteins, and likely spurious proteins, respectively. Clusters 1, 3, and 6 showed closer distances to known antigens, antibiotic targets, and virulence factors. Up to 21.8% of proteins in these clusters were structurally covered by modeling, which allowed assessment of druggability and discontinuous B-cell epitopes. Five proteins (4 in Cluster 1) were potential druggable targets for antibiotherapy. Eighteen proteins (11 in Cluster 6) were strong B-cell and T-cell immunogen candidates for vaccine development. Conclusively, we provide a feature-based schema to fractionate the functional dark proteome of critical pathogens for fundamental and biomedical purposes.

A theoretical analysis based on causal inference and single-instance learning

Although using single-instance learning methods to solve multi-instance problems has achieved excellent performance in many tasks, the reasons for this success still lack a rigorous theoretical explanation. In particular, the potential relation between the number of causal factors (also called causal instances) in a bag and the model performance is not transparent. The goal of our study is to use the causal relationship between instances and bags to enhance the interpretability of multi-instance learning. First, we provide a lower bound on the number of instances required to determine causal factors in a real multi-instance learning task. Then, we provide a lower bound on the single-instance learning loss function when testing instances and training instances follow the same distribution and extend this conclusion to the situation where the distribution changes. Thus, theoretically, we demonstrate that the number of causal factors in the bag is an important parameter that affects the performance of the model when using single-instance learning methods to solve multi-instance learning problems. Finally, combining with a specific classification task, we experimentally validate our theoretical analysis.