Sequence-based prediction of protein protein interaction using a deep-learning algorithm
Problem
Accurate high-throughput identification of protein-protein interactions is needed to understand protein function, disease occurrence and therapy design, but the robustness of existing computational methods on external datasets was unknown.
Why it matters
Protein-protein interactions underpin many biological processes, so reliable prediction aids disease understanding and therapy design.
What the AI did
A stacked autoencoder, a deep-learning algorithm, was used to predict protein-protein interactions from protein sequences and was evaluated on external datasets.
Results
97.19% average accuracy with 10-fold cross-validation; 87.99% to 99.21% accuracy on external datasets, superior to previous methods.
Impact
Peer-reviewed publication reporting superior accuracy over previous methods on external datasets.
Record
| Venue | BMC Bioinformatics |
|---|---|
| Published | 2017 |
| Work type | article |
| Authors | Tanlin Sun, Bo Zhou, Luhua Lai, Jianfeng Pei |
| Citations | 374 (OpenAlex, at import time) |
| References | 44 |
| Model or method | stacked autoencoder |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1186/s12859-017-1700-2 |
| Open access | gold |
| Keywords | protein-protein interaction, deep learning, stacked autoencoder, sequence-based prediction, cross-validation |
Importance rating
real but narrower advance
PPI prediction aids understanding of protein function, disease occurrence, and therapy design—broad biological relevance, but abstract states no scale of lives or industry affected.