Prediction of lncRNA-protein interacting pairs using LLM embeddings based on evolutionary information

Interactions of long non-coding RNAs (lncRNAs) with proteins is responsible for numerous cellular processes, including transcriptional regulation, chromatin remodeling, cell differentiation, and intracellular signaling. Several computational methods have been developed for predicting lncRNA–protein interaction (LPI) pairs; however, many existing approaches were developed using earlier versions of interaction databases, potentially limiting their applicability to more recently characterized interactions. In this study, we developed a sequence-based framework for LPI prediction using a large non-redundant dataset comprising 128,374 interacting and an equal number of non-interacting pairs. Initially, similarity-based approach BLAST was tried which had poor discriminative power, due to low sequence similarity between sequences. Subsequently, we developed convolutional neural network and machine learning models using sequence-derived features and protein language model embeddings. The CatBoost model integrating Dinucleotide Autocorrelation (DACC) features for lncRNAs and ESM-2-t30 embeddings for proteins achieved an AUC of 0.929 on the training dataset and 0.692 on the homology-aware validation dataset. These results demonstrate the potential of combining sequence-derived features and protein language model representations for LPI prediction while highlighting the challenges associated with generalization to sequences with limited homology to the training data. We developed standalone software and web server lncrnaPI for predicting LPI pairs, scanning lncRNA interacting proteins in proteome and protein interacting lncRNA in genomes ( https://webs.iiitd.edu.in/raghava/lncrnapi/ ).