Filtered circular fingerprints improve either prediction or runtime performance while retaining interpretability

Table 6 Pseudo-code for filtering circular fingerprints with supervised feature selection

\(^{(1)}\)Redundant := non-closed; i.e., a fragment is redundant if there exists a sub-fragment that matches the same compounds.
\(^{(2)}\)Fragments with the lowest p value of a \(\chi ^2\) test are removed; the test measures the correlation between the endpoint value distribution of all compounds, and the compounds that match the fragment

ISSN: 1758-2946