This website uses cookies to improve your experience. We'll assume you're ok with this, but you can opt-out if you wish. Privacy & Cookies policy
Challenge
Melanin-concentrating hormone receptor 1 (MCHR1), a GPCR target implicated in obesity, sleep disorders, depression, and anxiety, had no experimental structure available to guide a search for new antagonists. CACHE Challenge #5 asked 24 teams from twelve countries to predict up to 100 chemically novel, potent compounds active below 30 µM, working only from a shared set of 2,605 ChEMBL compounds and 753 patented molecules, while searching a chemical space of roughly 48 billion candidates. This solution brief shows how a purely ligand-based approach, without any target structure, can rival or beat structure-based and generative-AI methods on potency, novelty, and hit rate at this scale.
Solution
We used exaScreen®, our synthon-based virtual screening platform, to navigate the ~48 billion-compound Enamine REAL library guided entirely by three-dimensional molecular field similarity. The 753 known MCHR1 ligands were clustered into eight representative reference structures, and each candidate was represented by its QM-derived 3D molecular field — the electrostatic, hydrophobic, and steric pattern it presents to a binding site — generated by QaiM, our AI-based QM descriptor engine, at ultrafast speed. Candidate clusters were first prioritized by 2D similarity to the references, then refined with the more computationally intensive 3D field calculations; postprocessing based on 2D dissimilarity to all known MCHR1 ligands, combined with 3D field similarity, drove the chemical diversity of the hits. This staged workflow narrowed ~48 billion candidates to 38 compounds for synthesis and testing in Round 1, and was then used again in Round 2 to search for close 3D-field analogs of the two strongest Round 1 hits for lead optimization.
Result
Pharmacelera achieved the highest Top Score of all 24 participating teams, per CACHE’s official results, and the highest score on the blind Merged Selection task (89.9, more than 25 points ahead of the next-best team). From just 38 submitted compounds (2.6% of the challenge total), we delivered 4 of the 26 confirmed dose-response hits across the entire challenge, a 10% hit rate — about 1.5x the next-best team’s 6.7%. Round 2 lead optimization produced the challenge’s single most potent compound, CACHE5-HO_2127_49 (pKi 7.46, 34.7 nM), and Pharmacelera was the only team to identify four compounds with pKi greater than 7. That top compound also showed the greatest chemical novelty among all pKi > 6 submissions (Tanimoto distance 0.73 to the nearest known ligand), exceeding the diversity achieved by competing structure-based and generative-AI approaches. Together, these results demonstrate that exaScreen®’s ligand-based, physics-grounded workflow delivers potent, novel, and synthesizable hits at ultra-large scale — even with no experimental structure for the target.
Summary
In this case study we applied PharmScreen to explore a dataset of selected inhibitors for soluble Epoxide Hydrolase (sEH), a new target for relevant therapeutic strategy in Alzheimer Disease (AD). The data set contains 204 compounds (including hits and decoys), 41 of them identified as known inhibitors and 7 of them with a pIC50 > 9.4. The goal was to determine the maximum number of hits in the top ranked positions.
Solution
For the study, we analyzed the data set using 41 other various references. Among the different options of PharmScreen, the tool allows using different reference compounds in a single run, selecting the best ligand-reference similarity score (best similarity), and ranking the compounds in the library based on best similarities.
Result
Reference 33 displayed the highest similarity scores in the final ranking. A closer look at the results of this reference compound revealed 14 active compounds were ranked in top 10%, including 5 of the 7 more active compounds.
Inhibiting the function of HIVPR has proven to be a viable tactic in preventing the formation of new virions, thereby halting the infection propagated by the virus. This whitepaper delves into the effectiveness of screening tools in identifying potential inhibitors and opens up the exploration of new chemical series characterized by varied chemistry.
Solution
we harnessed the capabilities of PharmScreen® to spot potential hits with
innovative scaffolds for inhibiting HIVPR. The DUD-E+ dataset, comprising both hits and
decoys for HIVPR, was utilized. This specific version of the dataset for our screening
comprised a total of 36,286 compounds, with 536 being experimentally active (1.5%) and
35,750 being decoys (98.5%).
Result
We’ve spotlight PharmScreen® as an indispensable virtual screening
tool for discovering more diverse hits targeting HIVPR. The diversity explored by
PharmScreen® positions it as an ideal tool for the pursuit of new chemical series,
discovering innovative intellectual property, scaffold hopping, or backup compounds.
The effectiveness of the 3D hydrophobic descriptors is evidenced when contrasted with
the results garnered using 2D fingerprints. Although hit rate identification was comparable
for both methods, the chemical diversity of the compounds discovered by PharmScreen®
validates it as an excellent tool for identifying innovative and diverse compounds.
Challenge
Filling in the pipeline of antibiotics with true innovative starting points and approaches in Antimicrobial resistance (AMR), has been an issue for several years. From the perspective of CADD approaches, there are several challenges such as the limited knowledge about receptors, multiple additional factors affecting the final activity, or the desired molecular properties which are different from other therapeutic needs.
Solution
The application of ligand-based approaches could provide helpful information in the identification of new antimicrobials. In this case study, we applied PharmScreen, and a custom library for antibiotic research to identify inhibitors described in the literature for Mex flux pump protein (MFP). MC-207,110, a known inhibitor of MFP protein, was used as a reference.
Result
From the different compounds selected during the virtual screening, MBX3135 was chosen because of its properties. This compound is described as an inhibitor for AcrB, another multidrug efflux pump protein. Interestingly, MC-207,110 is also identified as an inhibitor of AcrB in literature, revealing a similar mechanism of action.
Challenge
A research group at a Big Pharma was looking to apply new technologies for fragment growing in their internal Fragment-Based Drug Discovery (FBDD) projects and compare PharmScreen’s capabilities compared to other existing tools in identifying new hits.
Solution
PharmScreen used 80 datasets provided by the client (each set composed of around 50K compounds) and only one hit and a reference fragment. Researchers evaluated the results delivered, and they observed that PharmScreen was able to retrieve the hit among the first 1,000 ranked (top 2%) molecules in 55 out of the 80 datasets.
Result
Challenge
Ligand-Based Virtual Screening (LBVS) experiments are an essential task in the early drug discovery stage. An ambitious aim in each experiment is to disclose active structures based on new scaffolds. However, most of the he descriptors used in LBVS are based on shape and atom connectivity. The application of these descriptors limits the identification of novel compounds as output.
Solution
In a case study on GPCRs, we validate the capability of PharmScreen, our LBVS tool, to explore alternatives chemotypes. PharmScreen is a strategy to compute 3D molecular similarity based on the molecular hydrophobicity derived from the quantum mechanical version of the Miertus − Scrocco − Tomasi (MST) continuum model.
Result
A set of 21 GPCR targets, which only include active molecules with an alternative scaffold regarding the reference structure, were compiled. The results point out the suitability of our MST-based hydrophobic descriptors for exploring a new chemical space. Considering the initial 5% for all the sets, 62 confirmed hits were reported by PharmScreen and 22 by another commercial tool, with focus on shape-based approaches.
Summary
Evotec performed a comparison between PharmScreen and a ligand- and shape-based tool from a different commercial vendor in terms of hit identification. The comparison between the two tools was executed by Evotec, using a challenging but real case. It was based on a known kinase inhibitor (reference compound) and two internal compound libraries that contain real activity data.
Solution
For the present comparison, Evotec used three similitude methods for the other tool. Dedicated scientist compared the scores with the default configuration of PharmScreen.
Evotec used its HTS library, with approximately 12,500 compounds. The library was divided in two data sets:
Result
When comparing the hits retrieved in the first 1000 positions, PharmScreen finds 40% and 36% of DS1 and DS2 hits, respectively, slightly better than 34% and 30% obtained by the other tool. Interestingly, 75% and 67% of these hits were unique to PharmScreen and not found by the other tool. This highlights the fact that each tool tested, explored a different chemical space.
Please get in touch and our team of experts will answer all your questions.