Abstract: The technology disclosed is directed to context-dependent base calling. The technology disclosed describes a system including memory storing k-mer-specific centroids for k-mers. The k-mer-specific centroids are learned by training a base calling pipeline to represent base calls of an already base called sequence in k-mer-specific time series, transform the k-mer-specific time series into predicted k-mer-specific centroids, merge the predicted k-mer-specific centroids on a sequencing cycle-by-sequencing cycle basis to generate predicted per-sequencing cycle intensity values, determine a training loss (e.g., a transformation loss) based on comparing the predicted per-sequencing cycle intensity values against known intensity values of the base calls, update the predicted k-mer-specific centroids based on the determined training loss, and store the updated centroids as the k-mer-specific centroids.

Abstract: The technology disclosed is directed to cluster segmentation and base calling. The technology disclosed describes a computer-implemented method including segmenting a population of clusters into a plurality of subpopulations of clusters based on one or more prior bases called at one or more prior sequencing cycles of a sequencing run. At a current sequencing cycle of the sequencing run, the method includes applying a mixture of four distributions to current sequenced data of each subpopulation of clusters in the plurality of subpopulations of clusters, the four distributions corresponding to four bases adenine (A), cytosine (C), guanine (G), and thymine (T), and the current sequenced data being generated at the current sequencing cycle. The method further includes base calling clusters in a particular subpopulation of clusters using a corresponding mixture of four distributions.