Aggregated Compound Biological Signatures Facilitate Phenotypic Drug Discovery and Target Elucidation

Cortés-Cabrera, Álvaro; Lucena‐Agell, Daniel; Redondo‐Horcajo, Mariano; Barasoaín, Isabel; Díaz, J. Fernando; Fasching, Bernhard; Petrone, Paula

doi:10.1021/acschembio.6b00358

Cited by 20 publications

(18 citation statements)

References 52 publications

(85 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Some previous work described methods where data from high-throughput screening assays stored on PubChem were aggregated to build compound biological fingerprints. 21 They were notably used for biological hit extension by phenotype similarity. However, that approach would not allow us to obtain information on previously uncharacterized compounds that are newly tested.…”

Section: Discussionmentioning

confidence: 99%

Compound Functional Prediction Using Multiple Unrelated Morphological Profiling Assays

et al. 2018

View full text Add to dashboard Cite

Phenotypic cell-based assays have proven to be efficient at discovering first-in-class therapeutic drugs mainly because they allow for scanning a wide spectrum of possible targets at once. However, despite compelling methodological advances, posterior identification of a compound's mechanism of action (MOA) has remained difficult and highly refractory to automated analyses. Methods such as the cell painting assay and multiplexing fluorescent dyes to reveal broadly relevant cellular components were recently suggested for MOA prediction. We demonstrated that adding fluorescent dyes to a single assay has limited impact on MOA prediction accuracy, as monitoring only the nuclei stain could reach compelling levels of accuracy. This observation suggested that multiplexed measurements are highly correlated and nuclei stain could possibly reflect the general state of the cell. We then hypothesized that combining unrelated and possibly simple cell-based assays could bring a solution that would be biologically and technically more relevant to predict a drug target than using a single assay multiplexing dyes. We show that such a combination of past screen data could rationally be reused in screening facilities to train an ensemble classifier to predict drug targets and prioritize a possibly large list of unknown compound hits at once.

show abstract

Section: Discussionmentioning

confidence: 99%

Compound Functional Prediction Using Multiple Unrelated Morphological Profiling Assays

et al. 2018

View full text Add to dashboard Cite

show abstract

“…11 Along similar lines, HTSFPs also contributed to different studies identifying the MoAs of drugs and natural products or identifying compounds with similar phenotypic profiles. [12][13] Multiple studies have shown that compounds with similar bioactivity profiles can have different structures. 6,11,[14][15] For instance, Petrone et al pointed out that compounds with similar HTSFPs have diverse chemical structures.…”

Section: Introductionmentioning

confidence: 99%

Application of Bioactivity Profile-Based Fingerprints for Building Machine Learning Models

Sturm

Sun

Vandriessche³

et al. 2018

J. Chem. Inf. Model.

View full text Add to dashboard Cite

The volume of high throughput screening data has considerably increased since the begin of the automated biochemical and cell-based assays era. This information rich data source provides tremendous repurposing opportunities for data mining. It was recently shown that biochemical or cell-based assay results can be compiled into so called high-throughput fingerprints (HTSFPs) as a new type of descriptor describing molecular bioactivity profiles which can be applied in virtual screening, iterative screening, and target deconvolution. However, so far studies around HTSFPs and machine learning have mainly focused on predicting the outcome of molecules in single highthroughput assays and no one has reported the modelling of compounds' biochemical assay activities towards a panel of target proteins. Therefore, there is a need for a detailed analysis of the performance of predictive models built with HTSFPs with respect to models built with more widely used structural descriptors both in terms of hit identification and of scaffold hopping potentials. In this article, in-house HTSFPs were built and combined with multi-task deep learning and support vector machine methods to build compound activity predictive models. Performances of HTSFP models were compared to the performances of models built with the conventional structural descriptors ECFPs. Moreover, we investigated the effect of high throughput screening false positives and negatives on the performance of deep learning models. Our results showed that the two fingerprints yielded in similar performances and in diverse hits with very little overlap, thus demonstrating the orthogonality of bioactivity profile based descriptors with structural descriptors. Therefore, modelling compound activity data using ECFP together with HTSFPs increases the scaffold potential of the predictive models.

show abstract

“…However, they have the advantage of being cheap, fast and nowadays quite accurate. This has been shown in retrospective studies for a broad range of molecules (Alvarsson et al., ; Keiser et al., ; Peón et al., ) and also in prospective studies in selected molecules (Cortes Cabrera et al., ; Keiser et al., ; Mugumbate et al., ).…”

Section: Introductionmentioning

confidence: 75%

MolTarPred: A web tool for comprehensive target prediction with reliability estimation

Peón

Ghislat

et al. 2019

Chem Biol Drug Des

View full text Add to dashboard Cite

Molecular target prediction can provide a starting point to understand the efficacy and side effects of phenotypic screening hits. Unfortunately, the vast majority of in silico target prediction methods are not available as web tools. Furthermore, these are limited in the number of targets that can be predicted, do not estimate which target predictions are more reliable and/or lack comprehensive retrospective validations. We present MolTarPred ( http://moltarpred.marseille.inserm.fr/), a user‐friendly web tool for predicting protein targets of small organic compounds. It is powered by a large knowledge base comprising 607,659 compounds and 4,553 macromolecular targets collected from the ChEMBL database. In about 1 min, the predicted targets for the supplied molecule will be listed in a table. The chemical structures of the query molecule and the most similar compounds annotated with the predicted target will also be shown to permit visual inspection and comparison. Practical examples of the use of MolTarPred are showcased. MolTarPred is a new resource for scientists that require a more complete knowledge of the polypharmacology of a molecule. The introduction of a reliability score constitutes an attractive functionality of MolTarPred, as it permits focusing experimental confirmatory tests on the most reliable predictions, which leads to higher prospective hit rates.

show abstract

Aggregated Compound Biological Signatures Facilitate Phenotypic Drug Discovery and Target Elucidation

Cited by 20 publications

References 52 publications

Compound Functional Prediction Using Multiple Unrelated Morphological Profiling Assays

Compound Functional Prediction Using Multiple Unrelated Morphological Profiling Assays

Application of Bioactivity Profile-Based Fingerprints for Building Machine Learning Models

MolTarPred: A web tool for comprehensive target prediction with reliability estimation

Contact Info

Product

Resources

About