csORF-finder: an effective ensemble learning framework for accurate identification of multi-species coding short open...

Area of Science:

• Genomics and Bioinformatics
• Molecular Biology
• Computational Biology

Background:

• Short open reading frames (sORFs) are nucleic acid fragments (<303 nt) potentially encoding small peptides.
• Translatable sORFs are found in mRNA untranslated regions and long non-coding RNAs (lncRNAs), impacting biological processes.
• Accurate computational tools are needed to identify translatable sORFs and discover novel functional peptides.

Purpose of the Study:

• To develop a highly accurate computational tool, csORF-finder, for differentiating coding sORFs (csORFs) from non-coding sORFs.
• To improve the prediction performance by introducing novel sequence-based features.
• To apply the tool for identifying potential csORFs in lncRNA datasets.

Main Methods:

• Designed ensemble models integrating Efficient-CapsNet and LightGBM (csORF-finder).
• Introduced a novel feature encoding scheme, trinucleotide deviation from expected mean (TDE), and computed in-frame sequence-based features.
• Benchmarked performance against state-of-the-art methods on multi-species datasets, including non-ATG initiation.

Main Results:

• The novel features significantly improved csORF-finder's performance compared to original features.
• csORF-finder outperformed existing methods in csORF prediction across multiple species.
• Applied csORF-finder to lncRNA datasets, generating a repository of potential csORFs for validation.

Conclusions:

• csORF-finder is a superior computational tool for accurate csORF identification.
• The tool facilitates high-throughput screening and functional characterization of peptides encoded by sORFs.
• The identified csORFs provide a valuable resource for experimental validation and further research.