Improving Classical Substructure-Based Virtual Screening to Handle Extrapolation Challenges

Abstract

Target-oriented substructure-based virtual screening (sSBVS) of molecules is a promising approach in drug discovery. Yet, there are doubts whether sSBVS is suitable also for extrapolation, that is, for detecting molecules that are very different from those used for training. Herein, we evaluate the predictive power of classic virtual screening methods, namely, similarity searching using Tanimoto coefficient (MTC) and Naive Bayes (NB). As could be expected, these classic methods perform better in interpolation than in extrapolation tasks. Consequently, to enhance the predictive ability for extrapolation tasks, we introduce the Shadow approach, in which inclusion relations between substructures are considered, as opposed to the classic sSBVS methods that assume independence between substructures. Specifically, we discard contributions from substructures included in ("shaded" by) others which are, in turn, included in the molecule of interest. Indeed, the Shadow classifier significantly outperforms both MTC (pValue = 3.1 × 10(-16)) and NB (pValue = 3.5 × 10(-9)) in detecting hits sharing low similarity with the training active molecules.

Database and Bioinformatics Studies of Probiotics

“…59,60 Intermediatesimilarity molecules have up to 30% chance of having the same activity in the same bioassay. 61 Remote similarity molecules have crosspharmacology relationships (i.e., active against different members of a protein family) but not necessarily the same activity in the same bioassay. 62 Therefore, the targets of the bacteriocins may in some cases be indicated by the knowledge of the targets of the structurally similar peptide drugs.…”

“…For instance, the modes of antimicrobial actions of the bacteriocins are the suppression of microbial growth by these antimicrobial peptides. − It is noted that microbial growth can also be inhibited by the approved peptide drugs that target the bacterial cell wall peptidoglycan sites. , Structural similarity frequently implies similar biological or therapeutic activities. High-similarity molecules typically have 30–81% chance of having the same activity in the same bioassay. , Intermediate-similarity molecules have up to 30% chance of having the same activity in the same bioassay . Remote similarity molecules have cross-pharmacology relationships (i.e., active against different members of a protein family) but not necessarily the same activity in the same bioassay .…”

CFam: a chemical families database based on iterative selection of functional seeds and seed-directed compound clustering

“…The criterion for assembling CFam family/families into a superfamily of intermediate to high similarity compounds is 2DF-TC >0.70, which was applied because compounds satisfying this criterion have been regarded as similar to one other ( 30 , 56 ) and those with slightly lower similarity typically have remote similarity ( 29 ). Compounds grouped by this intermediate-similarity criterion may have up to 30% chance of having the same activity in the same bioassay ( 11 ). These superfamilies were systematically named from the common target classes, chemical classes and individual family names of the constituent family names.…”

“…To make CFam chemical families more relevant to the applications in pharmaceutical, biomedical, agricultural, material and other industrial applications as well as to the research in chemistry and related scientific disciplines, the seeds of the CFam families were or are to be iteratively selected from hierarchically clustered approved drugs, clinical trial drugs, investigative drugs, bioactive molecules, human metabolites, food ingredients and additives, flavors and scents, agrochemicals, natural products, patented agents, toxic substances, purchasable compounds and other known compounds based on the literature-reported high-similarity measures ( 25 – 28 ). These families were further clustered into CFam superfamilies and classes by hierarchically clustering the seeds based on the literature-reported intermediate similarity ( 11 , 29 , 30 ) and remote similarity ( 3 , 13 , 30 ) measures. Although this iterative hierarchical clustering procedure seems similar to the incremental clustering algorithm used in selecting representative proteins for clustering proteins ( 31 ) and representative compounds for clustering large compound libraries ( 23 ), there are two significant differences.…”

Improved Protein-ligand Prediction Using Kernel Weighted Canonical Correlation Analysis

“…We can classify these methods into two: the struc-ture or docking approach and the ligand-based approach. Docking approaches make use of 3D structures of the chemical compounds or the proteins to find protein-ligand pairs which are more likely to bind [2], [4], [5]. On the other hand, ligand-based techniques usually employ machine learning algorithms in comparing known ligands and candidate ligands of a certain target even without any prior information regarding their structure [9], [14], [25].…”