The Jaccard similarity matrix of compound-protein pairs, where each element is Jaccard similarity of two sets S (C, P) and S (C′, P′), is also PD. 2. Let the minwise hashing matrix of compound-protein pairs be a matrix whose element is defined as the inner product of two compact fingerprints F (C, P) and F (C′, P′).
Each element of the pairwise kernel matrix of compound-protein pairs is defined as the number of common elements between two sets S (C, P) and S (C′, P′), i.e, | S (C, P) ∩ S (C′, P′)|.
However, the input of the SVM method in most previous works is the pairwise kernel similarity matrix of compound-protein pairs, which makes it difficult to analyze large-scale data.
To interrogate such data, MolClass generates a complete matrix of compound activities across many screens and thereby enables functional predictions for all molecules, even if not tested in a specific screen.
Computing the full similarity matrix of 4000 compounds, which is close to the size of larger fragment libraries, requires ca. 33 minutes.
In this work the authors integrated the experimental results from multiple databases into a bioactivity matrix of 52,498 compounds and 467 targets, including 246,088 observations.
This corresponds to 6% of the total of 103,360 possible compound receptor combinations in the full matrix of 3,230 compounds and 32 targets; see Table 2 for further details.
The bioactivity matrix of 157 compounds against 225 kinases is shown in Additional file 3: Figure S3 and given the importance of the data structure and density (as we will see later) this will be discussed here in some detail.
A total of 2769 unique structural fragments are found, and the fingerprints are represented as a matrix of 682 compounds × 2769 fragment descriptors.
More than 647,900 of these compounds were annotated for two or more protein targets, producing a well-populated matrix of inactive-compound annotations.
