Skip to main content

Target prediction, QSAR and conformal prediction


https://jcheminf.biomedcentral.com/articles/10.1186/s13321-018-0325-4 

You know that in the ChEMBL group, we love to play with the data we collect!! Back in April 2014, we started to work on a target prediction tool.  Wow! This was almost 5 years ago! Since then, we have continued to update the tool for each new ChEMBL release, providing you with the actual models and the result of the prediction on the ChEMBL website for the drug molecules. The good news is that these target predictions are not dead and a successor is on its way!

First, we would like to introduce you some closely related work. You may have heard about conformal prediction (CP). If not, it is a machine learning framework developed to associate confidence to predictions. I personally consider this as a requirement for decision making. Basically, you train a model as you would do in QSAR but then you first predict a so-called calibration set, for which you know the actual values. For each of these observations you obtain two probabilities: one for the active and one for the inactive class (in a typical classification scheme). Now that you have this information, each time you predict a new compound you compare its probabilities to those of the calibration set (the non-conformity scores as they are called) and you derived p-values for each class. Based on your predefined significance level, the compound can be assigned in different categories: only active or only inactive, but also both active and inactive or none of them. I am sure you can start seeing here the added value of CP!

Here I have briefly detailed how it works for classification models but CP can also be applied to regression models. If you want to know more about conformal prediction, I recommend you to read this book and also this very nice example of the application in drug discovery. Having learnt how to build conformal predictors, we were intrigued to know how well they perform against traditional QSAR models with our ChEMBL data!

With this in mind, we decided to build a panel of models using a substantial data set from ChEMBL. With our new protocol, we were able to build models for 788 targets (550 of them human targets). For the descriptors we used RDKit Morgan fingerprint (2048 bits and radius 2) and 6 physicochemical descriptors. For the machine learning part we used the good old Random Forests as implemented in Scikit-learn version 0.19. For the QSAR models, this is all that is needed, but for CP you need a framework and this was provided by the very nice library provided by Henrik Linusson.

The next part consisted of training the models and checking their internal performance, but we went a bit further and decided that with our models trained on ChEMBL_23 data, it would be interesting to see how they perform with new data in ChEMBL_24 in a so-called temporal validation. All the details, results and conclusion are presented in the recently accepted article!
Image result for right wrong decision

The dataset for each target is already available here and you can find the models ready to use there.

Feel free to take a look and to share your opinion in the comments.

Now, you remember that I started this post mentioning our good old target predictors. So does it mean a new generation of ChEMBL models using conformal prediction is ready to be launched for our users? Well, unfortunately not yet, but stay tuned!

Comments

Popular posts from this blog

Improvements in SureChEMBL's chemistry search and adoption of RDKit

    Dear SureChEMBL users, If you frequently rely on our "chemistry search" feature, today brings great news! We’ve recently implemented a major update that makes your search experience faster than ever. What's New? Last week, we upgraded our structure search engine by aligning it with the core code base used in ChEMBL . This update allows SureChEMBL to leverage our FPSim2 Python package , returning results in approximately one second. The similarity search relies on 256-bit RDKit -calculated ECFP4 fingerprints, and a single instance requires approximately 1 GB of RAM to run. SureChEMBL’s FPSim2 file is not currently available for download, but we are considering generating it periodicaly and have created it once for you to try in Google Colab ! For substructure searches, we now also use an RDKit -based solution via SubstructLibrary , which returns results several times faster than our previous implementation. Additionally, structure search results are now sorted by

ChEMBL brings drug bioactivity data to the Protein Data Bank in Europe

In the quest to develop new drugs, understanding the 3D structure of molecules is crucial. Resources like the Protein Data Bank in Europe (PDBe) and the Cambridge Structural Database (CSD) provide these 3D blueprints for many biological molecules. However, researchers also need to know how these molecules interact with their biological target – their bioactivity. ChEMBL is a treasure trove of bioactivity data for countless drug-like molecules. It tells us how strongly a molecule binds to a target, how it affects a biological process, and even how it might be metabolized. But here's the catch: while ChEMBL provides extensive information on a molecule's activity and cross references to other data sources, it doesn't always tell us if a 3D structure is available for a specific drug-target complex. This can be a roadblock for researchers who need that structural information to design effective drugs. Therefore, connecting ChEMBL data with resources like PDBe and CSD is essen

Improved querying for SureChEMBL

    Dear SureChEMBL users, Earlier this year we ran a survey to identify what you, the users, would like to see next in SureChEMBL. Thank you for offering your feedback! This gave us the opportunity to have some interesting discussions both internally and externally. While we can't publicly reveal precisely our plans for the coming months (everything will be delivered at the right time), we can at least say that improving the compound structure extraction quality is a priority. Unfortunately, the change won't happen overnight as reprocessing 167 millions patents takes a while. However, the good news is that the new generation of optical chemical structure recognition shows good performance, even for patent images! We hope we can share our results with you soon. So in the meantime, what are we doing? You may have noticed a few changes on the SureChEMBL main page. No more "Beta" flag since we consider the system to be stable enough (it does not mean that you will never

ChEMBL 34 is out!

We are delighted to announce the release of ChEMBL 34, which includes a full update to drug and clinical candidate drug data. This version of the database, prepared on 28/03/2024 contains:         2,431,025 compounds (of which 2,409,270 have mol files)         3,106,257 compound records (non-unique compounds)         20,772,701 activities         1,644,390 assays         15,598 targets         89,892 documents Data can be downloaded from the ChEMBL FTP site:  https://ftp.ebi.ac.uk/pub/databases/chembl/ChEMBLdb/releases/chembl_34/ Please see ChEMBL_34 release notes for full details of all changes in this release:  https://ftp.ebi.ac.uk/pub/databases/chembl/ChEMBLdb/releases/chembl_34/chembl_34_release_notes.txt New Data Sources European Medicines Agency (src_id = 66): European Medicines Agency's data correspond to EMA drugs prior to 20 January 2023 (excluding vaccines). 71 out of the 882 newly added EMA drugs are only authorised by EMA, rather than from other regulatory bodies e.g.

In search of the perfect assay description

Credit: Science biotech, CC BY-SA 4.0 Assays des cribe the experimental set-up when testing the activity of drug-like compounds against biological targets; they provide useful context for researchers interested in drug-target relationships. Ver sion 33 of ChEMBL contains 1.6 million diverse assays spanning ADMET, physicochemical, binding, functional and toxicity experiments. A set of well-defined and structured assay descriptions would be valuable for the drug discovery community, particularly for text mining and NLP projects. These would also support ChEMBL's ongoing efforts towards an  in vitro  assay classification. This Blog post will consider the features of the 'perfect' assay description and provide a guide for depositors on the submission of high quality data. ChEMBL's assays are typically structured with the overall aim, target, and method .  The ideal assay description is succinct but contains all the necessary information for easy interpretation by database u