basecallers

Basecallers are specialized software tools that convert raw sequencing signals into DNA sequences and per-base quality scores. They interpret instrument signals—such as electrical current traces from nanopore sequencing or fluorescence intensities over sequencing cycles—and assign a nucleotide (A, C, G, T, or N) to each position. The output is typically a read with quality information, commonly stored in FASTQ files.

Basecalling involves processing the raw signal, translating it into base calls, and calibrating or correcting for systematic noise. Early methods used statistical models such as hidden Markov models; modern basecallers increasingly rely on machine learning, including neural networks, often accelerated by GPUs. Some pipelines also include steps to correct for phasing, drift, or other artifacts that affect signal interpretation.

Platform notes and examples: for Oxford Nanopore Technologies, basecalling is a central step in turning squiggles into nucleotide sequences. Common basecallers include Guppy and Bonito, among others. PacBio SMRT sequencing uses a basecalling and signal-to-sequence pipeline within the SMRT Analysis framework (often referred to as part of pbbasecaller workflows) to convert polymerase kinetics and signal data into reads and consensus sequences. In Illumina and other short-read platforms, basecalling is typically integrated into the instrument software as a BCL-to-FASTQ conversion step (for example, using tools such as bcl2fastq or equivalent), and is sometimes described as the basecalling portion of the pipeline rather than a standalone tool.

Quality and impact: the accuracy of basecalling directly influences downstream analyses such as alignment, assembly, and variant calling. Ongoing improvements in modeling, including deep learning approaches, aim to reduce error rates, better handle modified bases, and mitigate platform-specific biases.