Deqformer: high-definition and scalable deep learning probe design method.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Target enrichment sequencing is crucial for genomic studies due to its cost-effectiveness and speed.
• Probe performance and uniform sequencing depth are critical for the success of these techniques.
• Accurate prediction of probe coverage depth is needed to optimize experimental design.

Purpose of the Study:

• To introduce Deqformer, a novel model for predicting probe coverage depth in target enrichment sequencing.
• To leverage oligonucleotide sequence information and advanced machine learning for accurate depth prediction.

Main Methods:

• Deqformer utilizes oligonucleotide sequences of probes, inspired by Watson-Crick base pairing.
• Two BERT encoders process forward and reverse probe strands, capturing sequence information.
• Encoded data are integrated with a feed-forward network for depth prediction.

Main Results:

• Deqformer achieved high accuracy (F3acc) on diverse datasets: 96.24% (SNP panel) and 99.66% (synthetic panel).
• Cross-dataset validation showed robust performance, with F3acc rates over 87.33% (lncRNA) and 72.56% (HD-Marker).
• The model effectively captures probe hybridization patterns for reliable predictions.

Conclusions:

• Deqformer offers a robust and accurate method for predicting probe coverage depth.
• The model provides a novel perspective for optimizing probe design in genomics.
• This approach can enhance the efficiency and effectiveness of target enrichment sequencing workflows.