Faster, Cheaper, Smarter: DBTL Cycles Transforming AI-Driven Antibody Discovery

A few weeks ago, I shared examples of how machine learning is enhancing antibody discovery and the challenges that come with its adoption. One major challenge that persists is low hit rates. Despite advances, de novo binders still achieve hit rates between 0.01% and 0.1%, particularly for challenging targets.

While we’ve made significant progress—considering that the number of possible 100-amino-acid protein sequences is 20^100 (far exceeding the number of atoms in the universe)—our current results are, at best, on par with traditional immunization methods. To fully unlock the potential of in silico discovery, hit rates need to improve by 100x.

Achieving this transformation requires two critical enablers: massive training datasets and scalable validation systems. While I’ll save the discussion on training data for another post, let’s focus on the latter. Faster DBTL (Design-Build-Test-Learn) cycles depend on scalable validation systems, yet current methods for generating and testing proteins remain slow, costly, and resource-intensive.

High-Throughput Technologies: Changing the Game

High-throughput platforms are stepping up to tackle these challenges. In my previous post, I highlighted how AI-driven discovery can automate and accelerate the build/test cycle by leveraging technologies such as DNA printers, high-throughput cell-free protein expression, and automated affinity measurements. These tools offer scalable solutions that are up to 10x faster than traditional immunization-based approaches.

For example, Biohat Biosciences recently revealed that their lab designs, synthesizes, and screens ~2,000 antibodies weekly—a feat made possible by extensive automation and significant capital investment. However, replicating such capabilities remains a challenging and time-consuming task.

Emerging high-throughput systems present newer, AI-first companies with an opportunity to bypass many of these hurdles, jumpstart data generation, and accelerate progress.

Breakthroughs in Efficiency

A recent publication from SPOC Bio caught my attention for its transformative approach that eliminates the need for protein purification entirely. Their platform starts with DNA libraries and uses cell-free expression systems to directly produce VHHs or protein binders in nanowells while simultaneously integrating affinity measurements.

Assuming the technology could scale, this innovation significantly reduces timelines, slashes costs, and increases throughput by up to 100x, making high-resolution SPR kinetic data generation far more efficient.

Could these advances extend to other proteins like soluble TCRs? Currently, producing a single soluble TCR costs over $1,000 and requires expert handling. High-throughput platforms could make this process scalable and cost-effective while generating the critical data needed to refine predictive models.

A Data-Driven Future for Biopharma

Data fuels AI, and the next wave of biopharma innovation will depend on scalable, accessible systems for data generation. From antibodies to TCRs, high-throughput technologies are paving the way for breakthroughs that were once unimaginable.

What other technologies could significantly reduce costs and timelines while increasing throughput?

Note: This is not a paid promotion, and I have no financial connections to the companies mentioned.

Publication: https://www.biorxiv.org/content/10.1101/2025.01.11.632576v1

Companies: SPOC Proteomics , Lydia Gushgari , BigHat Biosciences , Carterra