TExCNN: Predicting Gene Expression from DNA Sequences
Author Information
Author(s): Dong Guohao, Wu Yuqian, Huang Lan, Li Fei, Zhou Fengfeng
Primary Institution: Jilin University, Changchun, China
Hypothesis
Can pre-trained models improve the prediction of gene expression levels from genomic sequences?
Conclusion
The TExCNN framework significantly improves the accuracy of predicting gene expression levels from DNA sequences, especially with longer sequences.
Supporting Evidence
- TExCNN outperformed the state-of-the-art models with an average R2 score of 0.622.
- Extending the sequence length from 10,500 bp to 50,000 bp improved the average R2 score to 0.639.
- Incorporating additional biological features further enhanced prediction accuracy.
Takeaway
This study shows that using special models can help us guess how genes work better by looking at their DNA sequences.
Methodology
The study used a deep learning framework that integrates pre-trained models to generate word embeddings from DNA sequences for predicting gene expression.
Limitations
The study highlights limitations in traditional fine-tuning approaches and the computational demands of existing models.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website