Deep learning model predicts rice protein interactions from sequence data

Researchers have developed a deep learning framework that predicts protein-protein interactions in rice directly from amino acid sequences, offering a computational shortcut to mapping the molecular networks that underpin the crop’s metabolism and stress responses.

The model, called RCNNricePPI, was described in a peer-reviewed article published in the journal Amino Acids on 15 October 2026. The work was carried out by a team including Neeraj Budhlakoti, Nitesh Kumar Sharma, Sharanbasappa D Madival, Anshuman Singh, Sudhir Srivastava, Dwijesh Chandra Mishra, Mahender Kumar Singh, Atul Kumar and Girish Kumar Jha.

Protein-protein interactions are fundamental to cellular regulation, metabolism and stress-response pathways in rice, but the experimentally characterized interaction networks for the crop remain incomplete. The new tool is designed to prioritize candidate interactions for subsequent experimental investigation rather than to replace laboratory validation.

RCNNricePPI combines word2vec-derived amino acid embedding with a Siamese recurrent convolutional neural network that integrates convolutional layers with bidirectional gated recurrent units. In this architecture, pairs of protein sequences are processed in parallel branches, allowing the network to learn representations that capture similarities and differences between the two sequences.

The training data were curated from publicly available interaction resources, with the researchers removing self-interactions, redundant pairs and low-quality sequences. To reduce protein-level information leakage, model development and evaluation used a protein-aware partitioning strategy, including protein-disjoint five-fold cross-validation and an independent held-out test set.

At a classification threshold of 0.40, the model achieved an accuracy of 0.8996, a balanced accuracy of 0.7451 and an AUROC of 0.8295. A baseline logistic regression comparison reached values of 0.6459, 0.6964 and 0.7820 for the same metrics, respectively.

Performance was also assessed according to protein novelty. The model recorded AUROC values of 0.9259 for pairs where both proteins had been seen during training, 0.7995 for pairs with one unseen protein, and 0.7340 for pairs where both proteins were unseen. These figures indicate that prediction is strongest for familiar proteins and degrades as novelty increases.

Architectural ablation studies and comparisons with conventional machine learning approaches further supported the contribution of the integrated architecture, according to the authors’ reported results.

RCNNricePPI is implemented as an accessible web-based resource and is freely available at riceppi.biodb.in. The published version is an early-release, peer-reviewed accepted article that is citable and carries a permanent DOI, subject to further edits before the final Version of Record.

By narrowing the pool of candidate interactions for laboratory follow-up, the framework may help researchers studying how rice responds to environmental stress, though the reported results are computational performance measures and the authors position the tool as a starting point for experimental work rather than a validated source of biological findings.

Subject of Research: Deep learning prediction of protein-protein interactions in rice

Article Title: RCNNricePPI: leveraging deep learning to predict rice protein-protein interactions for climate-responsive agriculture

Article References: Budhlakoti, N., Sharma, N. K., Madival, S. D., Singh, A., Srivastava, S., Mishra, D. C., Singh, M. K., Kumar, A., & Jha, G. K. (2026). RCNNricePPI: leveraging deep learning to predict rice protein-protein interactions for climate-responsive agriculture. Amino Acids. https://doi.org/10.1007/s00726-026-03562-w

Image Credits: AI Generated

DOI: 10.1007/s00726-026-03562-w

Keywords: rice, protein-protein interactions, deep learning, RCNNricePPI, amino acid sequences, bioinformatics, machine learning, computational biology, stress response, Amino Acids journal, leveraging, deep