TargetCall: eliminating the wasted computation in basecalling via pre-basecalling filtering.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Basecalling converts nanopore sequencing signals to nucleotide reads, a critical but computationally expensive step.
• Current basecalling methods are resource-intensive, creating bottlenecks in genome analysis pipelines.
• Many sequencing reads are off-target and discarded later, representing wasted basecalling effort.

Purpose of the Study:

• To introduce TargetCall, the first pre-basecalling filter designed to reduce wasted computation.
• To eliminate off-target reads before the computationally expensive basecalling process.
• To improve the overall efficiency of nanopore sequencing data analysis.

Main Methods:

• TargetCall employs a two-component approach: LightCall and Similarity Check.
• LightCall, a lightweight neural network, generates preliminary, noisy reads.
• Similarity Check filters these noisy reads, classifying them as on-target or off-target against a reference genome.

Main Results:

• TargetCall enhances end-to-end basecalling runtime by up to $$ while preserving high recall for on-target reads.
• Downstream genomic analyses maintain high accuracy after TargetCall filtering.
• TargetCall outperforms prior methods in runtime, throughput, recall, precision, and generality.

Conclusions:

• TargetCall effectively reduces computational waste in nanopore sequencing by filtering reads pre-basecalling.
• The method significantly improves basecalling efficiency without compromising downstream analysis accuracy.
• TargetCall represents a substantial advancement in accelerating genomic data processing pipelines.