Predicting protein targets for drug-like compounds using transcriptomics

Abstract

An expanded chemical space is essential for improved identification of small molecules for emerging therapeutic targets. However, the identification of targets for novel compounds is biased towards the synthesis of known scaffolds that bind familiar protein families, limiting the exploration of chemical space. To change this paradigm, we validated a new pipeline that identifies small molecule-protein interactions and works even for compounds lacking similarity to known drugs. Based on differential mRNA profiles in multiple cell types exposed to drugs and in which gene knockdowns (KD) were conducted, we showed that drugs induce gene regulatory networks that correlate with those produced after silencing protein-coding genes. Next, we applied supervised machine learning to exploit drug-KD signature correlations and enriched our predictions using an orthogonal structure-based screen. As a proof-of-principle for this regimen, top-10/top-100 target prediction accuracies of 26% and 41%, respectively, were achieved on a validation of set 152 FDA-approved drugs and 3104 potential targets. We then predicted targets for 1680 compounds and validated chemical interactors with four targets that have proven difficult to chemically modulate, including non-covalent inhibitors of HRAS and KRAS. Importantly, drug-target interactions manifest as gene expression correlations between drug treatment and both target gene KD and KD of genes that act up- or down-stream of the target, even for relatively weak binders. These correlations provide new insights on the cellular response of disrupting protein interactions and highlight the complex genetic phenotypes of drug treatment. With further refinement, our pipeline may accelerate the identification and development of novel chemical classes by screening compound-target interactions.

Learning to Encode Cellular Responses to Systematic Perturbations with Deep Generative Models

“…Most drugs have their best performance achieved with VAE-learned latent representations rather than the raw expression profiles, and for five drugs, the best performance was achieved with the 12-signature-node-representation. Table 1 gives the best rank of the top known target for each drug, which is comparable to the Table 1 in (Pabon et al, 2018). Even though our approach is essentially an unsupervised learning method based purely on expression data, 13 out of 16 drugs received an equal or better rank than that from the random forest model trained with a combination of expression and protein-protein interaction features (Pabon et al, 2018) (bolded in Table 1).…”

“…Combining SMP and GP data can help establish connections between the MOAs of small molecules and genetic perturbations, which further help reveal the targets of small molecules (Lamb, 2007;Pabon et al, 2018). A simple approach is to examine whether a pair of perturbagens (a small molecule and a genetic perturbation) leads to similar transcriptomic profiles, or more intriguingly, similar latent representations that reflect the state of the cellular system ( Figure 6A).…”

“…A simple approach is to examine whether a pair of perturbagens (a small molecule and a genetic perturbation) leads to similar transcriptomic profiles, or more intriguingly, similar latent representations that reflect the state of the cellular system ( Figure 6A). Given a known pair of a drug and its target protein (Pabon et al, 2018) ( we created a ranked list of target genes knocked down in the GP samples, in a manner similar to an information retrieval task. We then compared different types of representations to identify which were more effective (measured as top and mean ranking, see Methods) for retrieving target genes than others.…”

“…While there are numerous studies using LINCS data to investigate the mechanism-of-action (MOA) of drugs and to promote clinical translation of MOA information (Iwata et al, 2017;Lamb et al, 2006;Pabon et al, 2018;Siavelis et al, 2015;Subramanian et al, 2017;Wang et al, 2016), few studies aim to use the data to learn to represent the cellular signaling system as an information encoder and examine how different perturbagens affect the system. It can be imagined that perturbing different signaling components at different levels of a signaling cascade would lead to compositional statistical structure in the data.…”

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

“…Liu et al [10] performed comparative analysis of genes frequently regulated by drugs based on connectivity map transcriptome data. Pabon et al [11] predicted protein targets for drug-like compounds using transcriptomics.…”