SPOC x AI
10-100x Challenge
Can AI deliver breakthrough improvements in drug design, outperforming existing technologies? This Challenge is designed to provide definitive validation.
We will focus on two in-clinic antibody drugs, each previously optimized using state-of-the-art methods. Participants are invited to apply AI models, trained on high-resolution SPR kinetic datasets provided at no cost, to further enhance binding affinity by >10X or improve pH-responsiveness.
Can AI deliver breakthrough improvements in drug design, outperforming existing technologies? This Challenge is designed to provide definitive validation.
We will focus on two in-clinic antibody drugs, each previously optimized using state-of-the-art methods. Participants are invited to apply AI models, trained on high-resolution SPR kinetic datasets provided at no cost, to further enhance binding affinity by >10X or improve pH-responsiveness.
The SPOC × AI Challenge aims to demonstrate the AI community’s ability to improve prior-optimized antibody sequences using models trained on high-resolution SPR datasets to optimize binding features across therapeutic modalities. Can AI enhance binding affinity by 10–100X? Can AI design engineer pH-responsiveness to reduce binding at physiological pH while increasing affinity in acidic tumor microenvironments (lower pH)?
The SPOC × AI Challenge aims to demonstrate the AI community’s ability to improve prior-optimized antibody sequences using models trained on high-resolution SPR datasets to optimize binding features across therapeutic modalities. Can AI enhance binding affinity by 10–100X? Can AI design engineer pH-responsiveness to reduce binding at physiological pH while increasing affinity in acidic tumor microenvironments (lower pH)?
The Challenge will consist of two iterative rounds, with separate tracks focused on optimizing distinct parameters (see below). The SPOC team will select two FDA-approved in-clinic monoclonal antibody drugs (in scFv, VHH, or Fab formats) targeting well-characterized antigens.
In Round 1, single-residue deep mutational scanning (DMS) will be performed across CDRs and their flanking residues, substituting each position with alanine and six other amino acids. A DMS ~1,000-variant scFv library will be synthesized on the SPOC SPR chip and screened against target antigens to generate high-resolution kinetic data (RU, ka, kd, KD, t₁/₂) under multiple pH conditions. Developability attributes of variants such as stability, aggregation, immunogenicity, and viscosity will be assumed constant for the purpose of this Challenge.
SPOC Proteomics will provide the datasets at no cost to registered participants for AI/ML model training. Participating teams may introduce unlimited combination mutations in CDRs or framework regions, as long as binding to the original epitope is retained. Each team may submit up to 5 sequences per target per track (max 20 sequences total per participant). Submitted variants will be synthesized on SPR chips, screened against targets, and ranked based on Round 1 performance.
To support further model training and guide Round 2 submissions, comprehensive Round 1 results (SPR datasets) will be shared with participants. Round 2 variants will be synthesized and screened on SPOC SPR chips, with performance evaluated across both challenge tracks. Top sequences from Round 2 will be ranked by affinity and pH responsiveness tracks, winners announced, and prize certificates issued. To note - initial DMS data sets, comprehensive data sets from both round 1 and 2 results will be shared with participating teams. Only 25 top ranking sequences will be displayed publicly. Teams can choose to participate anonymously.
To receive contest updates please use the button above to register and follow us on LinkedIn. If you’d like to meet with us regarding a general interest in SPOC or custom project, please reach out to info@spoc.bio.
Let’s solve the challenges of data generation and validation for AI-enabled drug discovery together!
In this challenge, on the date submissions open, SPOC Proteomics will provide all registered participants with a downloadable dataset containing all kinetic parameters (RU, ka, kd, KD, t1/2) for mutationally scanned anti-HER2 VHH and Trastuzumab (in scFv form) binding to HER2. Data will include every amino acid substituted at each CDR position and the two positions flanking each CDR on either side. Participants will be tasked to use this data to train their model of choice and generate new sequences with the goal of improving the KD. Participants will have one month to generate and submit their sequences to our portal, at which time the SPOC team will synthesize the submitted sequences and produce the scFv/VHH on our SPOC SPR chips to generate kinetic data for HER2 binding. After Round 1 data collection, a second data drop will occur where teams can use the data from Round 1 to further improve and submit additional sequences for Round 2. After Round 2, winners will be announced and prizes awarded.
To receive contest updates please use the button above to register and follow us on LinkedIn. If you’d like to meet with us regarding a general interest in SPOC or custom project, please reach out to info@spoc.bio.
Let’s solve the challenges of data generation and validation for AI-enabled drug discovery together!
SPOC Proteomics’ transformative platform generates sub-pico-molar resolution kinetic data (RU, ka, kd, KD, t1/2) for 384 up to 1,000 antibody-fragment sequences (scFv or Fab or VHH) simultaneously using SPR (see our pre-print here). Using our proprietary direct-from-DNA cell-free protein synthesis and capture technology (see our Nature Communications Biology publication here), we deliver kinetic data at a fraction of cost and time compared to current methods.
SPOC Proteomics’ transformative platform generates sub-pico-molar resolution kinetic data (RU, ka, kd, KD, t1/2) for 384 up to 1,000 antibody-fragment sequences (scFv or Fab or VHH) simultaneously using SPR (see our pre-print here). Using our proprietary direct-from-DNA cell-free protein synthesis and capture technology (see our Nature Communications Biology publication here), we deliver kinetic data at a fraction of cost and time compared to current methods.
Timeline
Timeline
Round 1
Round 1
May 2025: Registration open.
July: antibody CDR DMS kinetic data sets shared with registered participants.
August: Deadline to submit AI-designed sequences.
September: Top 10 sequences in each track from Round 1 posted to public leaderboard. Full SPR data sets from AI-designed round 1 sequences shared with participating teams.
SPOC SPR data for training made available to registered participants only. Teams can participate in one track, or all three tracks listed below.
May 2025: Registration open.
July: antibody CDR DMS kinetic data sets shared with registered participants.
August: Deadline to submit AI-designed sequences.
September: Top 10 sequences in each track from Round 1 posted to public leaderboard. Full SPR data sets from AI-designed round 1 sequences shared with participating teams.
SPOC SPR data for training made available to registered participants only. Teams can participate in one track, or all three tracks listed below.
April 2025: Registration open.
June: antibody CDR DMS kinetic data sets shared with registered participants.
July: Deadline to submit AI-designed sequences.
August:
• SPOC SPR data sets from AI-designed round 1 sequences shared with participating teams.Top 10 sequences posted to public leaderboard.
SPOC SPR data for training made available to only registered participants. Teams can participate in one track, or all three tracks listed below.
Round 2
Round 2
September: Sequence submissions open after Round 1 data drop to participants.
October: Deadline to submit improved AI-designed Round 2 sequences, trained on new data sets.
November:Winners announced for each track, each target. Top 25 sequences posted to public leaderboard, Prize certificates distributed
2026: Manuscript co-authored with winning participants
Participation in round 1 is required for round 2 submissions. A participant can win only one prize per track, for each target.
September: Sequence submissions open after Round 1 data drop to participants.
October: Deadline to submit improved AI-designed Round 2 sequences, trained on new data sets.
November:Winners announced for each track, each target. Top 25 sequences posted to public leaderboard, Prize certificates distributed
2026: Manuscript co-authored with winning participants
Participation in round 1 is required for round 2 submissions. A participant can win only one prize per track, for each target.
Two Track Competition and Prizes
Two Track Competition and Prizes
01
01
Antibody sequences optimized for highest affinity (lowest KD value)
Antibody sequences optimized for highest affinity (lowest KD value)
02
02
Antibody sequences tuned for conditional pH activity, with higher binding affinity at acidic pH 6 but lower affinity at pH 7.4 (measured here as ratio of affinities)
Antibody sequences tuned for conditional pH activity, with higher binding affinity at acidic pH 6 but lower affinity at pH 7.4 (measured here as ratio of affinities)
Prizes for each Track ( total 9)
Prizes for each Track ( total 9)
192 sequence order free (includes DNA to data)
60% off of 384 sequence order
40% off of 384 sequence order
In addition, each winner will have a chance to be a co-author on a journal paper featuring the dataset generated in the contest.
192 sequence order free (includes DNA to data)
60% off of 384 sequence order
40% off of 384 sequence order
In addition, each winner will have a chance to be a co-author on a journal paper featuring the dataset generated in the contest.
Sample Data
Sample Data
Above is a partial dataset from an anti-HER2 VHH where all three CDRs were mutationally scanned with 4 amino acid substitutions at each position and kinetic values generated (KD, ka, kd, t1/2, and Rmax are shown). N/A indicates loss of binding of the VHH-variant to HER2 – where Rmax was below the threshold cutoff for the assay and therefore kinetic values cannot be assessed. HaloTag is on the c-terminus of protein-of-interest (VHH in this case) and is expressed last. It is a dehalogenase enzyme that forms covalent bond with chloro-alkane functionalization on the SPR biosensor surface. Anti-HaloTag assay indicates in-frame expression of protein-of-interest followed by HaloTag protein, and proper folding of HaloTag enzyme. HaloTag Rmax is an indicator that the sequence expressed as designed and was captured onto the SPOC chip, specifying whether undetectable Target binding is due to lack of expression or loss of binding.
Such comprehensive datasets from antibody DMS libraries screened with respective targets will be provided to teams as an excel table, for AI training, as described in the contest overview.
Above is a partial dataset from an anti-HER2 VHH where all three CDRs were mutationally scanned with 4 amino acid substitutions at each position and kinetic values generated (KD, ka, kd, t1/2, and Rmax are shown). N/A indicates loss of binding of the VHH-variant to HER2 – where Rmax was below the threshold cutoff for the assay and therefore kinetic values cannot be assessed. HaloTag is on the c-terminus of protein-of-interest (VHH in this case) and is expressed last. It is a dehalogenase enzyme that forms covalent bond with chloro-alkane functionalization on the SPR biosensor surface. Anti-HaloTag assay indicates in-frame expression of protein-of-interest followed by HaloTag protein, and proper folding of HaloTag enzyme. HaloTag Rmax is an indicator that the sequence expressed as designed and was captured onto the SPOC chip, specifying whether undetectable Target binding is due to lack of expression or loss of binding.
Such comprehensive datasets from antibody DMS libraries screened with respective targets will be provided to teams as an excel table, for AI training, as described in the contest overview.
SPOC routinely measures pico-molar affinity interactions as shown by the above anti-TNFα VHH produced on SPOC chip binding to recombinant TNFα, and the below p53 protein produced on SPOC chip binding to an anti-p53 antibody. The data resolution generated by SPOC SPR chips is sufficient to detect sub-pico-molar binders with confidence. Furthermore, SPOC SPR screening can be applied to resolve and characterize femto-molar affinity binding interactions using the chaser method, as detailed by John G. Quinn, Analytical Chemistry 2025 97 (11), 10.1021/acs.analchem.4c06023
SPOC routinely measures pico-molar affinity interactions as shown by the above anti-TNFα VHH produced on SPOC chip binding to recombinant TNFα, and the below p53 protein produced on SPOC chip binding to an anti-p53 antibody. The data resolution generated by SPOC SPR chips is sufficient to detect sub-pico-molar binders with confidence. Furthermore, SPOC SPR screening can be applied to resolve and characterize femto-molar affinity binding interactions using the chaser method, as detailed by John G. Quinn, Analytical Chemistry 2025 97 (11), 10.1021/acs.analchem.4c06023
Proteins and scFvs/VHH are produced at the time of need directly on SPOC chips using our proprietary cell-free lysate production and protein capture technology. The above data for p53 proteins produced on SPOC chip binding
to an anti-p53 antibody in solution demonstrate that even if expression and capture levels differ, the resolved kinetic parameters are the same as long as capture levels (Rmax) are above the assay cutoff threshold. This demonstrates the reliability of kinetic data generated on SPOC SPR chips.
Proteins and scFvs/VHH are produced at the time of need directly on SPOC chips using our proprietary cell-free lysate production and protein capture technology. The above data for p53 proteins produced on SPOC chip binding
to an anti-p53 antibody in solution demonstrate that even if expression and capture levels differ, the resolved kinetic parameters are the same as long as capture levels (Rmax) are above the assay cutoff threshold. This demonstrates the reliability of kinetic data generated on SPOC SPR chips.
SPOC x AI 10-100x Challenge
SPOC x AI 10-100x Challenge
Registration
Registration
For more information on the SPOC platform, get in touch:
Information
Contact Us
Privacy & Conditions
1600 Adams Drive Suite236 Menlo Park, CA 94025
7201 E Henkel Way, Suite 285 Scottsdale, AZ 85255
480-219-9506
All rights reserved © 2024
For more information on the SPOC platform, get in touch:
Information
Contact Us
Privacy & Conditions
1600 Adams Drive Suite236 Menlo Park, CA 94025
7201 E Henkel Way, Suite 285 Scottsdale, AZ 85255
480-219-9506
All rights reserved © 2024
For more information on the SPOC platform, get in touch:
Information
Contact Us
1600 Adams Drive
Suite 236
Menlo Park, CA 94025
7201 E Henkel Way
Suite 285
Scottsdale, AZ 85255
480-219-9506
For more information on the SPOC platform, get in touch:
Information
Contact Us
1600 Adams Drive
Suite 236
Menlo Park, CA 94025
7201 E Henkel Way
Suite 285
Scottsdale, AZ 85255
480-219-9506