Extending the small molecule similarity principle to all levels of biology

Abstract

We present the Chemical Checker (CC), a resource that provides processed, harmonized and integrated bioactivity data on 800,000 small molecules. The CC divides data into five levels of increasing complexity, ranging from the chemical properties of compounds to their clinical outcomes. In between, it considers targets, off-targets, perturbed biological networks and several cell-based assays such as gene expression, growth inhibition and morphological profilings. In the CC, bioactivity data are expressed in a vector format, which naturally extends the notion of chemical similarity between compounds to similarities between bioactivity signatures of different kinds. We show how CC signatures can boost the performance of drug discovery tasks that typically capitalize on chemical descriptors, including target identification and library characterization. Moreover, we demonstrate and experimentally validate that CC signatures can be used to reverse and mimic biological signatures of disease models and genetic perturbations, options that are otherwise impossible using chemical information alone. an organic molecule (A: Chemistry) that interacts with one or several protein receptors (B: Targets), triggering perturbations of biological pathways (C: Networks) and eliciting phenotypic outcomes that can be measured in e.g. cell-based assays (D: Cells) before delivery to patients (E: Clinics). Using these five categories, we classified the information stored in major compound databases, including chemogenomics resources, cell-based screens and, when available, clinical reports of drug effects (Methods).We then divided each level (A-E) into five sublevels (1-5) corresponding to distinct types or scopes of the data. In total, the CC contains 25 well-defined categories meant to illustrate the most relevant aspects of small molecule characterization. In particular, we stored the 2D (A1) and 3D (A2) structures of compounds, together with their scaffolds (A3), functional groups (A4) and physicochemistry (A5). We also retrieved therapeutic targets (B1) and drug metabolizing enzymes (B2), and molecules co-crystallized with protein chains (B3). We fetched literature binding data (B4) from major chemogenomics databases, and high-throughput target screening results (B5). Moving to a higher order of biology, we looked for ontological classifications of compounds (C1) and focused on human metabolites in a genomescale metabolic network (C2). In addition, we kept the pathways (C3), biological processes (C4) and protein-protein interactions (C5) of the previously collected binding data. To capture cell-level information, we gathered differential gene expression profiles (D1) and compound growth-inhibition potencies across cancer cell lines (D2). Similarly, we gathered sensitivity profiles over an array of yeast mutants (chemical genetics) (D3), as well as cell morphology changes (high-content screening) (D4). Additional cell sensitivity data available from the literature were also collected (D5). To organize clinical data, we used the traditio...

Identification and drug-induced reversion of molecular signatures of Alzheimer’s disease onset and progression inAppNL-G-F,AppNL-Fand 3xTg-AD mouse models

“…Next, we sought to identify small molecules with the capacity to “revert” the transcriptional traits of AD mice, potentially ameliorating the disease phenotype. As a chemical space, we considered the over 800,000 bioactive compounds included in the Chemical Checker (CC) ( 25 ). The CC provides different types of bioactivity descriptors (a.k.a.…”

“…We can then use these signatures to connect small molecules to desired outcomes observed in phenotypic experiments such as genetic perturbation assays, as popularized in the context of transcriptomics by the Connectivity Map initiative (61). Indeed, we proved that CC signatures could be used to identify compounds able to revert the transcriptional changes induced by AD mutations (PSEN1 M146V and APP V717F ) in SH-SY5Y cells (25). We now explored whether the capacity to prioritize compounds that revert molecular signatures could be translated to in vivo models, in which the phenotypic effects of this reversion can be measured.…”

“…Recently, we extended the idea of using small molecules to mimic or revert biological signatures beyond transcriptional profiles and demonstrated the capacity of our approach to identify compounds able to revert specific expression alterations of AD genes (e.g., BIN1, GRIND2D, etc.) in APP V717F and PSEN1 M146V SH-SY5Y cells ( 25 ). However, despite its potential, this approach to revert global signatures has been validated only in cell cultures and thus its in vivo relevance is still uncertain.…”

An ensemble learning approach for modeling the systems biology of drug-induced injury

“…We tried two different strategies that led to different results: (i) focusing on samples from a unique cell line and dose-time point for each drug, and (ii) selecting the most correlated samples for each drug. This is by no means comprehensive and various possible strategies such as using other cell lines and dose-time points, or discarding the samples with low correlation between replicas ('distil_cc_q75' < 0.2) and selecting the sample with highest transcriptional activity score [43] could be investigated further.…”

Universal nature of drug treatment responses in drug-tissue-wide model-animal experiments using tensor decomposition-based unsupervised feature extraction

“…In order to consider the gene expression profile for the drug design, some data bases are established. For example, chemical checker [1] includes gene expression for computer aided drug design while PharmacoDB [2] is fully implemented to consider dose dependence of drug treated cell lines for drug design. Many papers were published to make use of gene expression profiles for computer aided drug design [3,4].…”