Search

Protein-protein interactions (PPIs) are important for many biological processes, but predicting them from sequence data remains challenging. Existing deep learning models often cannot generalize to proteins not present in the training set and do not provide uncertainty estimates for their predictions. To address these limitations, we present TUnA, a Transformer-based uncertainty-aware model for PPI prediction. TUnA uses ESM-2 embeddings with Transformer encoders and incorporates a Spectral-normalized Neural Gaussian Process. TUnA achieves state-of-the-art performance and, importantly, evaluates uncertainty for unseen sequences. We demonstrate that TUnAs uncertainty estimates can effectively identify the most reliable predictions, significantly reducing false positives. This capability is crucial in bridging the gap between computational predictions and experimental validation.

Protein-protein interactions (PPIs) are of key importance for understanding how cells and organisms function. Thus, in recent decades, many approaches have been developed for the identification and discovery of such interactions. These approaches addressed the problem of PPI identification either by an experimental point of view or by a computational one. Here, we present an updated version of UniReD, a computational prediction tool which takes advantage of biomedical literature aiming to extract documented, already published protein associations and predict undocumented ones. The usefulness of this computational tool has been previously evaluated by experimentally validating predicted interactions and by benchmarking it against public databases of experimentally validated PPIs. In its updated form, UniReD allows the user to provide a list of proteins of known implication in, e.g., a particular disease, as well as another list of proteins that are potentially associated with the proteins of the first list. UniReD then automatically analyzes both lists and ranks the proteins of the second list by their association with the proteins of the first list, thus serving as a potential biomarker discovery/validation tool.

Protein chains are generally long and consist of multiple domains. Domains are the basic of elements of protein structures that can exist, evolve and function independently. The accurate and reliable identification of protein domains and their interactions has very important impacts in several protein research areas. The accurate prediction of protein domains is a fundamental stage in both experimental and computational proteomics. The knowledge is an initial stage of protein tertiary structure prediction which can give insight into the way in which protein works. The knowledge of domains is also useful in classifying the proteins, understanding their structures, functions and evolution, and predicting protein-protein interactions (PPI). However, predicting structural domains within proteins is a challenging task in computational biology. A promising direction of domain prediction is detecting inter-domain linkers and then predicting the reigns of the protein sequence in which the structural domains are located accordingly. Protein-protein interactions occur at almost every level of cell function. The identification of interaction among proteins and their associated domains provide a global picture of cellular functions and biological processes. It is also an essential step in the construction of PPI networks for human and other organisms. PPI prediction has been considered as a promising alternative to the traditional drug design techniques. The identification of possible viral-host protein interaction can lead to a better understanding of infection mechanisms and, in turn, to the development of several medication drugs and treatment optimization. In this work, a compact and accurate approach for inter-domain linker prediction is developed based solely on protein primary structure information. Then, inter-domain linker knowledge is used in predicting structural domains and detecting PPI. The research work in this dissertation can be summarized in

Protein chains are generally long and consist of multiple domains. Domains are the basic elements of protein structures that can exist, evolve, and function independently. The accurate and reliable identification of protein domains and their interactions has very important impacts in several protein research areas. The accurate prediction of protein domains is a fundamental stage in both experimental and computational proteomics. The knowledge of domains is an initial stage of protein tertiary structure prediction which can give insight into the way in which proteins work. The knowledge of domains is also useful in classifying proteins, understanding their structures, functions and evolution, and predicting proteinprotein interactions (PPI). However, predicting structural domains within proteins is a challenging task in computational biology. A promising direction of domain prediction is detecting inter-domain linkers and then predicting the reigns of the protein sequence in which the structural domains are located accordingly. Protein-protein interactions occur at almost every level of cell function. The identification of interaction among proteins and their associated domains provide a global picture of cellular functions and biological processes. It is also an essential step in the construction of PPI networks for human and other organisms. PPI prediction has been considered as a promising alternative to the traditional drug design techniques. The identification of possible viral-host protein interactions can lead to a better understanding of infection mechanisms and, in turn, to the development of several medication drugs and treatment optimization. In this work, a compact and accurate approach for inter-domain linker prediction is developed based solely on protein primary structure information. Then, inter-domain linker knowledge is used in predicting structural domains and detecting PPI. The research work in this dissertation can be summarized in three main

Maps of protein-protein interactions (PPIs) are essential to uncover cellular processes and metabolic processes in a cell. However, various high-throughput biological experiments are time-consuming and labor-intensive, resulting in interactions of high false positive and false negative rates. The fact that most interaction networks remain sparse and incomplete motivates scientists to develop computational methods to predict protein-protein interactions accurately and automatically. However, state-of-the-art prediction algorithms cannot make satisfactory predictions. In this paper, we propose a simple yet effective approach SimpleTrPPI, to improve the accuracy of predicting protein-protein interactions in the target PPI network with the aid of another source PPI network. We attempt to transfer and borrow useful knowledge from the source PPI network using similarities of protein nodes between two protein interaction networks. Similarities are computed taking both protein sequence similarities and topological structures of protein networks into account. Two protein-protein interaction networks, Helicobacter pylori (target network) and Human (source network), are used to verify the feasibility of our proposed method. Our experimental results show that SimpleTrPPI can achieve more than 5% accuracy improvement compared to the baseline methods. ©2010 IEEE.

Maps of protein-protein interactions (PPIs) are essential to uncover cellular processes and metabolic processes in a cell. However, various high-throughput biological experiments are time-consuming and labor-intensive, resulting in interactions of high false positive and false negative rates. The fact that most interaction networks remain sparse and incomplete motivates scientists to develop computational methods to predict protein-protein interactions accurately and automatically. However, state-of-the-art prediction algorithms cannot make satisfactory predictions. In this paper, we propose a simple yet effective approach SimpleTrPPI, to improve the accuracy of predicting protein-protein interactions in the target PPI network with the aid of another source PPI network. We attempt to transfer and borrow useful knowledge from the source PPI network using similarities of protein nodes between two protein interaction networks. Similarities are computed taking both protein sequence similarities and topological structures of protein networks into account. Two protein-protein interaction networks, Helicobacter pylori (target network) and Human (source network), are used to verify the feasibility of our proposed method. Our experimental results show that SimpleTrPPI can achieve more than 5% accuracy improvement compared to the baseline methods. ©2010 IEEE.