Join the CACHE Challenges!

We are pleased to launch the DREAM x CACHE Target 2035 Drug Discovery Challenge, which is now accepting predictions. 

Part of the Target 2035 initiative, this successor to the inaugural DREAM Target 2035 Drug Discovery Challenge is made possible via a collaboration between The Structural Genomics Consortium (SGC), the labs of Damian Young at Baylor College of Medicine and Neelagandan Kamariah at InStem, IBM (DREAM) and Sage Bionetworks, with logistical and technical support from Conscience and BEACON. All data and results will be shared openly and publicly. 

The Scientific Challenge

In this DEL-ML challenge, participants are invited to retrospectively and prospectively predict small molecules that bind phosphoglycerate kinase 2 (PGK2), an enzyme essential for sperm motility and a promising target for non-hormonal contraception.

 Participants have access to all chemical structures and PGK2 binding data from a DNA-encoded library (DEL) screen of close to a billion compounds, on which they can train their machine learning (ML) models. They should then use their models to retrospectively retrieve hits from an affinity selection mass spectrometry (ASMS) screen of 400K compounds that were confirmed in a kinase assay. Participants who retrieve the most diverse set of hidden hits will then be invited to prospectively predict novel hits from Enamine’s catalog, which will be procured and tested experimentally. 

The winners of the challenge will be determined based on the number of novel chemical series identified from the test set and the overall number of hits, as determined in the prospective phase of the challenge.  

Participation Cost

Participation in the retrospective phase of the challenge is free.

The top 10 participants from the retrospective phase will be invited to advance to the prospective phase where they will each purchase 50 compounds in-stock at Enamine (total cost < $1000). The organizers  will cover 100% of the costs of testing compounds experimentally.

Participants from low and medium income countries who advance to the prospective phase can have all expenses covered by the SGC if they are part of the MAINFRAME Open Science Machine Learning Network for Drug Discovery. Organizers will also cover the costs of procuring compounds for the top 4 participants from the retrospective phase if they are members of MAINFRAME. 
 

Canadian participants click here for details on available funding.

How to Participate

Sign up to participate via the Synapse platform. The first submission deadline for predictions in the retrospective phase of the challenge is September 15th. To learn more about the challenge and ask any questions you might have, please register to attend the challenge launch webinar, which will take place via Zoom on July 7, 9am EDT/ 3PM CEST.

The eighth CACHE challenge is focused on GID4, the substrate-binding unit of the CTLH ubiquitin ligase complex. Compounds binding GID4 can lead to targeted protein degradation strategies that highjack CTLH E3 ligase activity for therapeutic applications across disease areas.

Participants are asked to find chemically novel molecules occupying the substrate binding pocket of GID4, occupied by a peptide in structure 6CCT and by a potent and selective chemical probe in structure 7SLZ (see Figure).

The seventh CACHE challenge is focused on PGK2, a kinase and a potential contraceptive drug target effecting sperm bioenergetics that is being actively pursued within the Gates Foundation non-hormonal contraceptive program. This challenge is supported by unpublished data provided by program members Damian Young and Kevin MacKenzie (BCM), Neelagandan Kamariah (inStem), Levon Halabelian and Hui Peng (SGC-Toronto), and Tim Willson (SGC-UNC).

CACHE participants are tasked with identifying orthosteric inhibitors that are selective for PGK2 over its close homologue PGK1, a glycolytic enzyme with available crystal structures in the PDB. All sidechains in the ligand binding pocket are conserved except for Y242 (PGK2) vs F242 (PGK1) and A255 (PGK2) vs T255 (PGK1) at the edge of the pocket.

Figure 1: Structural chemistry of PGK1/2 inhibition. The ATP-bound structure has the PDB identifier 2X15 (a). Overview of structures (b) and ligands (c), most unpublished, that may inform the computational prediction of non-quinazoline inhibitors selective for PGK2 vs PGK1.

Available structural data include ATP-bound and ligand-bound hPGK1, ATP bound and apo mPGK2, and hPGK2 bound to two selective PGK2 quinazoline inhibitors (compound 21 and compound 47). hPGK1 bound to a selective quinazoline PGK1 inhibitor (compound 45) is also provided. Atomic coordinates and electron density files are available here.

Preliminary data on unpublished non-quinazoline ligands with some selectivity for PGK2 are shown (Figure 1c).

CACHE participants are also provided with 1388 hPGK2 hit candidates from a DNA-encoded library (DEL) screen (5 reads or more) that are not enriched in a hPGK1 screen. These compounds have not been tested off-DNA. As typical for DEL data, the signal-to-noise ratio is expected to be low and should be interpreted accordingly (ex: Montoya et al 2025).

The quinazoline series is well-advanced and will be published soon. CACHE #7 participants are therefore asked to discover SELECTIVE hPGK2 inhibitors that DO NOT contain a quinazoline.

The sixth CACHE challenge is focused on SETDB1, a multi domain protein involved in epigenetic mechanisms and an immuno-oncology target. Participants are asked to find molecules occupying one or multiple subcavities of the histone binding groove of the SETDB1 triple Tudor domain (TTD). A few ligands (PDB codes 7CJT, 8UWP, 6AU3) occupy one or two of three subcavities that accommodate methylated or acetylated lysine sidechains, as shown in this structure overview.

The fifth CACHE challenge is focused on MCHR1, a G-Protein coupled receptor linked to obesity, sleep disorders and with potential roles in depression and anxiety disorders.

CACHE participants are asked to predict chemically novel antagonists for MCHR1.

More details are shared in the data package on the Challenge #5 page.  

This CACHE challenge is focused on the TKB domain of CBLB, an E3 ubiquitin-protein ligase with hundreds of compounds reported in the patent literature from multiple organizations, all chemically related. CACHE participants are asked to predict compounds with new scaffolds.

More details are shared in the data package on the Challenge #4 page.

Predict ligands that bind to the ADPr site of SARS-CoV-2 Nsp3 macrodomain (Mac1). 

Mac1 interferes with the immune response to viral infection and is a valid target against SARS-CoV.

More details are shared in the data package on the Challenge #3 page.

Finding ligands targeting the conserved RNA binding site of SARS-CoV-2 NSP13.

The RNA-binding site of NSP13 is the most conserved site in SARS-CoV-2 and across coronaviruses. The underlying selective pressure is expected to apply to future coronavirus outbreaks and drugs binding this site should be effective against future pandemic threats.

More details are shared in the data package on the Challenge #2 page.

Predict hits for the WD40 repeat (WDR) domain of LRRK2, a Parkinson's Disease target.

LRRK2 is the most commonly mutated gene in familial Parkinson's Disease. 

More details are shared in the data package on the Challenge #1 page.