Fig. 2

Graphical representation of decOM. Our method preprocesses an input k-mer matrix of aOral metagenomic samples and its possible contaminants, divides it into sinks and sources and then estimates and outputs the proportions of each source environment in the sink. The core idea in the classification step is that if a k-mer is present in the sink s represented by the vector \(\textbf{m}^{(s)}\), and in the source vector \(\textbf{m}^{(j)}\) with environment label \(l_j\), then a ball is added to the bin with label \(l_j\) (Ex: K-mer AAACG is present in the input sink S and in source S1 labelled as skin, S5 labelled as aOral and S7 labelled as mOral; hence, one ball is added to the bin of skin, aOral and mOral respectively). After every entry in the sink vector is compared against every entry of every vector in the sources, decOM outputs the estimated environment proportions and the hard label assigned to the sink s is that of the environment with the highest contribution