A new AI/ML approach for ligand identification
Only 1% of compound to protein interactions are currently known. We are on a mission to predict the remaining 99%.
A needle in a haystack
“We ran a drug discovery assay but did not find promising hits” is something we have already heard many times. Running an assay to find promising candidates is actually like finding a needle in a haystack. It can be an expensive and lengthy effort as such assays are not cheap, can be difficult to set-up or sometimes expensive to run at a great scale.
In order to define what could be the promising compounds, scientists usually start from existing knowledge, and look for existing hits in the literature. This is unfortunately not always a straightforward process. In many cases, literature does not contain information about hits, or only a few hits are known. This can be the case of novel targets or targets associated with rare diseases. Also, even if literature does contain information about hits, this information is by nature biased by what is known or what is disclosed .
Of course, in silico approaches exist. An obvious one is docking. Docking is a method which predicts the preferred orientation of one molecule to a second when a ligand and a target are bound to each other to form a stable complex. Even if this approach is useful, it is computationally very expensive, and as such, extremely slow. Another approach is advanced knowledge research, through Natural Language Processing or knowledge graphs, for example. These approaches are interesting but they will be only limited to what has been previously published, or base future findings on what already exists. In that way, they will also be biased, or at least, limited.
With Zeptomics, we use Artificial Intelligence to go beyond what is known. We propose a radically new approach to drug discovery. A cognitive technology which is fast, bias free, and unblinds drug discovery.
Unleashing the possibilities of drug discovery
A cognitive model
Our vision for Zeptomics was to build a model that can help fill the gap of scientific knowledge, and predict promising hits for any therapeutic target or any protein. In other words: we wanted to be able to just ask Zeptomics to help us fill the gaps of science by discovering the exact biophysical reasons why compounds bind to macromolecules, and in doing so, be able to predict which compound will bind to which protein of interest.
This ambition is now well underway: Zeptomics is able to discover, on its own, correlations between any therapeutic target and any compound, in a manner that is both fast and precise.
How is it possible?
Well, Zeptomics regroups a slew of innovations.
Within our internal AI lab, we have conceived, tested and implemented novel ways of using AI for drug discovery.
Zeptomics has been trained on millions of highly curated and selected data points from literature and other data sources.
Zeptomics is regularly trained. We constantly feed new data to Zeptomics so it regularly becomes more and more intelligent.
If we speak in this post about hit identification, Zeptomics already goes much beyond. Step by step, it is on its way to virtualize the whole drug discovery pipeline.
Unique performance figures
In order to predict hit compounds for a protein, Zeptomics only needs the sequence of this protein. In a few minutes, it will provide a complete list of probable hits.
Because it is cognitive, Zeptomics is able to work both on targets for which data is available and targets for which no data is available.
Hit rate for targets for which prior hits are known
On average, for every 2 compounds recommended by Zeptomics, 1 will be a hit.
Hit rate for targets for which no prior hits are known
On average, for every 4 compounds recommended by Zeptomics, 1 will be a hit.
What does it mean for drug hunters
You can see Zeptomics as a very smart filter that performs thousands of tests in silico and identifies both the opportunities and the risks along the drug discovery pipeline.
The teams we partner with use Zeptomics to find on-target hits, but not only.
Thanks to its wide spectrum, Zeptomics also predicts off-target hits, meaning it can decrease the risk of adverse events.
Zeptomics goes even further: it can filter compounds that are synthesizable, or with low off-target effects, or with acceptable ADMET properties.
And this, in a few hours only.
An engine for good
We use Zeptomics to unblind and unrisk the whole discovery process for diseases for which there is no cure or therapy.
We are on our way to build a virtual drug discovery pipeline that considerably reduces the risk of clinical failure and accelerates discovery efforts, in close contact with drug hunters.
If you would like to use Zeptomics to accelerate, get in touch through our contact form. If now is not the time, subscribe to our newsletter