Deep imputation on large-scale drug discovery data
OA paper outlining the practical applications of deep imputation on large-scale drug discovery data. It compares deep learning to traditional QSAR methods.
When evaluating any new technology, it is important to establish how you will validate whether it will deliver a return on investment (ROI). This sounds obvious and straightforward, but it isn’t always that easy. In this blog, we will cover the process of Cerella evaluation.
When validating method performance, we often rely on model statistics. Indeed, we have validated Cerella in many previous publications that show robust statistics and give confidence that Cerella is building accurate and reliable models (read more here). If we rely on the model statistics alone, however, we often miss the true value that such methods can provide in real-life prospective applications.
One way to demonstrate Cerella’s true power is through active learning. By utilising imputation and multi-parameter optimisation, we will illustrate how we can move from focusing on model statistics to looking at the return on investment Cerella can provide by shortening the drug discovery process.
What makes Cerella valuable?
Many of us are familiar with the expensive, time-consuming and challenging nature of a typical drug discovery or compound optimisation process. Starting with a sparse data set where a subset of compounds has been measured in a selection of the early-stage experiments and some in more costly experiments, making decisions on what to measure or progress next isn’t always easy
With the help of Cerella, we can make confident decisions to guide our project in the right direction and make the most of the resources available. Cerella learns from the limited data available and imputes the missing values in our large data matrix, providing us with a complete set of predictions without running experiments in the lab – saving time, money and experimental resources. We can use the predictions in our multi-parameter scoring profile to identify our best compounds to prioritise and take forward for experimental confirmation.
How to demonstrate the value of Cerella in a short pilot project?
Translating such value into a monetary figure in prospective applications is often difficult because there is no baseline without Cerella with which to compare. Retrospective analysis is the easiest way to show Cerella’s return on investment by comparing the progress of a project in practice with what can be achieved with Cerella’s guidance.
Therefore, we can apply Cerella to a project that has already been completed in the labs. We can model the data available at the start of the project and use it to make predictions for the later compounds that have not been ‘seen’ by the model. From here, we can select the compounds that are predicted to be best to ‘synthesise’. Cerella can also identify the most informative compounds to add to the model, along with the most informative measurements across the new compounds to improve the model’s ability to make predictions.
By adding the selected compounds and data, we can then rebuild the model in an iterative active learning cycle, selecting new compounds and data at each iteration.
To validate success, we can track the number of compounds tested and the number of experimental measurements added to find the best compounds identified by the project in practice, and compare these with the real-time project progression without Cerella. For example, if Cerella selected X% of the compounds synthesised in practice during the project and used Y% of the available data, we can then turn this into a total percentage saved to achieve the same result. We have also been able to highlight additional interesting opportunities that weren’t explored in the real-time project.
Looking at an active learning application we can turn stats into $$ saved and demonstrate the value provided by Cerella in finding the best compounds in a project quickly, with reduced synthetic and experimental efforts.
Expectations for your Cerella evaluation
Want to see how Cerella could support your organisation? Each evaluation is different depending on your goals and objectives. However, to ensure you get maximum value from your evaluation, there’s a few things we need from you.
Data requirements
To get set up for success, a data set with a minimum of 1000 compounds and multiple experimental endpoints representing the activities and properties that are relevant to your projects. The data set should be curated with one compound per row (see data for cheminformatics blog for guidelines on how to achieve this) and a way to identify the chronological order of compounds synthesised or tested in the project.
We will work with you to define a scoring profile for multi-parameter optimisation to identify high-quality compounds for your project. This will define the properties and criteria for selecting compounds during the active learning process.
Structure of the evaluation
Typically, an evaluation lasts 3 months. Throughout the whole process, you will be supported by our research scientists, who will provide guidance and training while you become familiar with Cerella.
The evaluation will start with a meeting to discuss your data and project objectives and agree on the specific criteria for evaluating Cerella’s performance.
Our data scientists will configure Cerella for your data and monitor the model-building process for your initial model. They will then report the validation results to confirm that everything is working as expected.
We’ll then provide training on Cerella, so you can learn how best to harness its value, and schedule regular meetings throughout the evaluation to discuss progress, present results, and continue training.
Want to evaluate Cerella?
Get in touch using the form below and a member of the team will be in touch to discuss your needs.
Ready to transform to AI-guided drug discovery?
Simply complete the form and our team will be happy to chat about how Cerella could improve your discovery workflows.
More Cerella resources
How is AI transforming drug discovery chemistry?
Imagine you’re trying to find the correct key to unlock a treasure box, but there are billions of keys to…
How much do AI drug discovery platforms cost?
Which AI platform do you need? The first thing you’ll need to do is decide what you need to achieve…
