Protein (Multi-)Location Prediction: Using Location Inter-Dependencies in a Probabilistic Framework

TL;DR

This paper introduces a probabilistic framework using Bayesian network classifiers to improve the prediction of multiple protein locations by directly modeling inter-dependencies among locations, outperforming previous methods.

Contribution

The paper presents a novel method that incorporates location inter-dependencies into multi-location protein prediction, moving beyond independent or limited combination-based approaches.

Findings

Incorporating inter-dependencies significantly improves prediction accuracy.

The system performs comparably to top methods like YLoc+ on multi-localized proteins.

The approach allows predictions beyond training set location combinations.

Abstract

Knowing the location of a protein within the cell is important for understanding its function, role in biological processes, and potential use as a drug target. Much progress has been made in developing computational methods that predict single locations for proteins, assuming that proteins localize to a single location. However, it has been shown that proteins localize to multiple locations. While a few recent systems have attempted to predict multiple locations of proteins, they typically treat locations as independent or capture inter-dependencies by treating each locations-combination present in the training set as an individual location-class. We present a new method and a preliminary system we have developed that directly incorporates inter-dependencies among locations into the multiple-location-prediction process, using a collection of Bayesian network classifiers. We evaluate our system on a dataset of single- and multi-localized proteins. Our results, obtained by incorporating inter-dependencies are significantly higher than those obtained by classifiers that do not use inter-dependencies. The performance of our system on multi-localized proteins is comparable to a top performing system (YLoc+), without restricting predictions to be based only on location-combinations present in the training set.