Protein-Coding sORFs Prediction Based on U-Net and Coordinate Attention with Hybrid Encoding

Small proteins encoded by small open reading frames (sORFs) exhibit significant biological activity in crucial biological processes such as embryonic development and metabolism. Accurately predicting whether sORFs encode small proteins is a key challenge in current research. To address this challenge, many methods have been proposed, however, existing methods rely solely on biological features as the sequence encoding scheme, which results in high feature extraction complexity and limited applicability across species. To tackle this issue, we proposed a deep learning architecture UAsORFs based on hybrid coding of sORFs sequences. In contrast to mainstream prediction methods, this framework processes sORF sequences using a mixed encoding approach, including both one-hot and gapped k-mer encodings, which effectively captures global and local sequence information. Additionally, it autonomously learns to extract features of sORFs and captures both long-range and short-range interactions between sequences through U-Net and coordinate attention mechanisms. Our research demonstrates significant progress in predicting encoded peptides from eukaryotic and prokaryotic sORFs, particularly in improving the cross-species predictive MCC index on the eukaryotic dataset.

@article{Wang2024,
  title     = {Protein-Coding sORFs Prediction Based on U-Net and Coordinate Attention with Hybrid Encoding},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {15},
  number    = {8},
  year      = {2024},
  publisher = {The Science and Information Organization},
  author    = {Ziling Wang and Wenxi Yang and Zhijian Qu},
  doi       = {10.14569/IJACSA.2024.0150856},
  url       = {https://doi.org/10.14569/IJACSA.2024.0150856}
}