Mappability of drug-like space: towards a polypharmacologically competent map of drug-relevant compounds

Abstract

Intuitive, visual rendering—mapping—of high-dimensional chemical spaces (CS), is an important topic in chemoinformatics. Such maps were so far dedicated to specific compound collections—either limited series of known activities, or large, even exhaustive enumerations of molecules, but without associated property data. Typically, they were challenged to answer some classification problem with respect to those same molecules, admired for their aesthetical virtues and then forgotten—because they were set-specific constructs. This work wishes to address the question whether a general, compound set-independent map can be generated, and the claim of “universality” quantitatively justified, with respect to all the structure–activity information available so far—or, more realistically, an exploitable but significant fraction thereof. The “universal” CS map is expected to project molecules from the initial CS into a lower-dimensional space that is neighborhood behavior-compliant with respect to a large panel of ligand properties. Such map should be able to discriminate actives from inactives, or even support quantitative neighborhood-based, parameter-free property prediction (regression) models, for a wide panel of targets and target families. It should be polypharmacologically competent, without requiring any target-specific parameter fitting. This work describes an evolutionary growth procedure of such maps, based on generative topographic mapping, followed by the validation of their polypharmacological competence. Validation was achieved with respect to a maximum of exploitable structure–activity information, covering all of Homo sapiens proteins of the ChEMBL database, antiparasitic and antiviral data, etc. Five evolved maps satisfactorily solved hundreds of activity-based ligand classification challenges for targets, and even in vivo properties independent from training data. They also stood chemogenomics-related challenges, as cumulated responsibility vectors obtained by mapping of target-specific ligand collections were shown to represent validated target descriptors, complying with currently accepted target classification in biology. Therefore, they represent, in our opinion, a robust and well documented answer to the key question “What is a good CS map?”

Graphical Abstract

Appendix: ChEMBL data curation protocol

This “Appendix” describes the data curation protocol concerning Challenges 3 and 4, as shown in Fig. 14 and explained in further details below.

Fig. 14

Workflow of data mining and curation for external Homo sapiens-target related validation. For mined targets, compounds were labeled as active or inactive according to their reported activity type: inhibition %, K_i, IC50, EC50, “potency”—and a chosen activity threshold (AT). Targets that have only potency activity type associated entered only challenge 4, others entered both challenges

First, the complete list of biological targets of Homo sapiens was retrieved from ChEMBL. For each target labeled as “single protein” (2474 proteins) associated ligand activity data were uploaded using a script available as Supporting Information. Analysis and binning of activity data was done as follows:

• 1. Compounds associated with an inhibition percentage equal or less than 50 % are labeled as inactives.

• 2. Compounds with an associated dose-dependent activity value lower than an imposed activity threshold are labeled as actives.

• 3. Compounds with an associated dose-dependent activity value equal or higher than ten times the activity threshold were labeled as inactives.

Eventually, compounds with multiple entries leading to contradictory tentative activity class assignments were ignored. Compounds for which label active or inactive could not be set were also ignored.

In the steps 2 and 3, several activity thresholds were tested: 1000, 500, 100 and 50 nM. A threshold was used for a given target if:

• more than 100 compounds could be successfully classified for the given target, and

• out of these, at least 20 compounds are active, and

• inactives are representing more than 50 % of the set.

If the above rules could not be satisfied, the target was discarded. If more than one threshold value satisfied these conditions, the resulting dataset with the proportion of active compounds closest to 25 % was kept.

If the activity value used in the previous steps was the K_i, IC50 or EC50, the targets entered the both Challenge 3 and Challenge 4. If the activity type was potency, the associated target entered the Challenge 4 only. Those data are results of high-throughput screening (HTS) campaigns.

Supporting Information features two tar archives, Challenge 3.tar.gz and Challenge 4.tar.gz respectively, containing files named <ChEMBLTarget_ID>_S<subset number>.class which report, for the named target, a list of ligand ChEMBL IDs next to their attributed class (1-inactive/2-active). Distributions contain also the files target.chid_AT_name, listing on three columns the validated target ChEMBL IDs, associated activity thresholds AT (in nM) and the name of the target as given in ChEMBL database.