Predicting gene and protein expression levels from DNA and protein sequences with Perceiver

Area of Science:

• Computational Biology
• Genomics
• Proteomics

Background:

• Gene and protein expression are fundamental to organism function.
• Accurate prediction of mRNA and protein levels is vital for scientific research.
• Current mRNA prediction models (CNN, LSTM) have limitations in handling sequence dependencies.

Purpose of the Study:

• To develop novel models for predicting mRNA and protein levels.
• To leverage Transformer-based architectures for improved sequence analysis.
• To address limitations in existing methods for gene and protein level prediction.

Main Methods:

• Utilized a Perceiver model with a Transformer-based architecture and attention module.
• Developed DNAPerceiver for mRNA level prediction from DNA sequences.
• Developed ProteinPerceiver for protein level prediction from protein sequences.
• Developed Protein&DNAPerceiver integrating both DNA and protein sequences.

Main Results:

• Evaluated models on diverse datasets including cell lines, mice, and cancer tissues.
• Demonstrated the effectiveness of Perceiver-type models in predicting mRNA and protein levels.
• The proposed models successfully capture long-range interactions within biological sequences.

Conclusions:

• Introduced a novel Perceiver architecture for accurate mRNA and protein level prediction.
• Future work may enhance predictions by incorporating regulatory and epigenetic information.
• Source code is available for public use.