Pan-microalgal dark proteome mapping via interpretable deep learning and synthetic chimeras.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Microalgal genomes harbor a substantial "dark proteome" of unclassified sequences.
• Conventional tools struggle to identify and classify these homologous sequences.

Purpose of the Study:

• To develop an AI-powered framework for classifying translated ORFeomes in microalgae.
• To address the challenge of the "dark proteome" and enhance proteomic data annotation.

Main Methods:

• Utilized transformer and state-space models for sequence representation (LA4SR).
• Trained on approximately 77 million algal amino acid sequences across ten phyla.
• Evaluated classification accuracy, recall, and inference speed compared to traditional methods like BLASTP+.

Main Results:

• LA4SR achieved near-complete recall and accelerated classification by over 10,700×.
• The model demonstrated robust generalization to unseen sequences, requiring minimal training data.
• Terminal Information-free (TI-free) sequence models maintained high accuracy and enhanced scalability.

Conclusions:

• LA4SR provides a highly efficient and accurate method for interrogating the microbial dark proteome.
• The framework integrates biological context with computational innovation for broad disciplinary accessibility.
• LA4SR facilitates deeper insights into algal evolutionary and biophysical features through interpretable sequence patterns.