Skip to main content
Fig. 1 | BMC Bioinformatics

Fig. 1

From: Multi-task learning sparse group lasso: a method for quantifying antigenicity of influenza A(H1N1) virus using mutations and variations in glycosylation of Hemagglutinin

Fig. 1

Workflow of multi-task learning sparse group LASSO (MTL-SGL). The framework was composed of three levels: data, model, and prediction levels. To avoid bias from sample sizes, training data were divided into several subsets according to their data source, data type, and antigenic clusters. The pairwise genetic difference matrix and pairwise antigenic distance vector were generated from hemagglutination inhibition (HI) assay data and hemagglutinin subunit HA1 sequence data in the data level for each dataset. The MTL-SGL model selected antigenicity-associated features by analyzing correlations between genotype and phenotype by minimizing an objective function. In the model level, each selected feature is given a numeric weight as its impact. Antigenic distances among new viruses can be inferred from their HA1 sequences in the prediction level by weights from the model level. Sequence-based maps are generated by using a multidimensional scaling algorithm from pairwise antigenic distances inferred from HA1 sequences. Min, minimum; miss, missing. Antigenic cluster: BE95, A/Beijing/262/1995(H1N1)-like virus; BR07, A/Beijing/262/1995(H1N1)-like virus; NC99, A/New Caledonia/20/1999(H1N1)-like virus; pdm09, A(H1N1)pdm09-like virus; pdm1918, 1918 pandemic H1N1-like virus; RU77, A/USSR/90/1977(H1N1)-like virus; SG86, A/Singapore/6/1986(H1N1)-like virus; SI06, A/Solomon Islands/3/2006(H1N1)-like virus

Back to article page