De novo antibody design has advanced rapidly, with high hit rates enabling direct binder characterization without requiring high-throughput screening. Chai's latest models design drug-like biologics against challenging, therapeutically relevant targets, achieving high hit rates, sub-atomic structural precision, and developability profiles on par with clinical antibodies using at most 96 designs per target. We illustrate the frontiers this unlocks by designing functional antibodies mediating GPCR agonism, and highly specific antibodies discriminating tumor-specific neoepitopes among other applications. Beyond reducing the cost and timelines of large screening campaigns, in silico design is opening new horizons for creative, targeted therapeutics addressing unmet clinical needs.

Nabla Bio's de novo design platform generates developable, high-affinity antibodies faster than conventional wet-lab workflows can evaluate them, demanding characterization that can keep pace. We pair AI-driven design with in-depth wet-lab characterization at scale to assess how de novo designs measure up across the properties that matter most for therapeutic development—developability, specificity, and manufacturability—in both monospecific and multispecific formats.

Traditionally, Design of Experiments (DOE) has been constrained by manual execution and disconnected workflows. We present an integrated framework that combines automated experimental design, digital workflow translation, and continuous DOE–execute–analyze–refine cycles for protein production optimization. Applied to CHO protein expression, this approach improves productivity, quality, and robustness, and demonstrates a practical path toward autonomous, system-level optimization in biologics development.

Increasingly complex biologic molecules, combined with an already highly competitive drug discovery landscape, are challenging protein production groups to develop innovative strategies that keep pace with accelerated timelines while maintaining the quality and control required for transition into candidate development. We present a high throughput, miniaturized platform for rapid, parallel evaluation of biologic candidates leveraging site specific integration for stable CHO cell line generation. This engineered process provides early insights and flexibility for seamless alignment across discovery and into development.

Efficient recovery of multispecific biologics requires optimized engineering and purification processes. We developed an integrated small-scale purification and analytical QC platform enabling in-process prediction of purification strategies for complex modalities. Upstream of wet-lab validation, we deploy AI/ML models to predict molecular attributes from sequence, including expression titer and chain pairing. These approaches link sequence design, assembly prediction, and purification strategy selection to accelerate lead triaging and delivery of complex formats.

Despite the vital role that proteins play in biotechnology, recombinant protein production is more art than science, relying on laborious trial-and-error expression optimization instead of rational predictions. To address this gap, we train predictive, machine learning–based models on high-quality protein expression datasets to develop the cell-free protein synthesis (CFPS) based equivalent of a protein “printer." For any sequence provided by a user, the latter should allow rapid and reliable protein synthesis. We will report on our progress towards this vision.

The proliferation of AI has made it easier than ever to design proteins for improved activity, stability, or other characteristics. However, there is a bottleneck bringing these designs into the real world. Comprehensive integration of AI in areas like drug development, protein engineering and mutagenesis studies demand an equally rapid downstream screening platform. Opera Bioscience has developed a high-throughput protein screening platform harnessing a bacterial secretion host to produce high-purity recombinant proteins. Opera incorporates automation-capable purification screens and analytical workflows to enable rapid optimization from screening to scaling protein manufacturing.

Voltage-gated calcium (CaV) and sodium (NaV) channels are essential regulators of cellular excitability but remain challenging to study because of their structural complexity and multi-subunit organization. We developed a modular platform using cell-free protein expression coupled with co-translational nanodisc assembly to reconstruct functional human ion channel systems without living cells. The platform enables rapid synthesis, membrane insertion, and functional characterization of CaV and NaV channels in defined lipid environments, supporting analysis of channel regulation, pharmacological modulation, and signaling networks. This synthetic approach provides a scalable foundation for mechanistic studies, high-throughput drug screening, medical countermeasure development, and programmable bioelectronic technologies.

Despite rapid advances in AI-driven protein design, a critical gap remains between predictive models and experimental reality. This talk presents an actionable framework for integrating experimental datasets with sequence-and structure-based predictive tools to guide designs for proteins, binding domains, and CAR T constructs. Case studies will highlight how combining experimental and computational datasets improves insights, outcomes and decision-making.

Better protein expression models require data that is clean, quantitative, and learnable. Flock Bio builds high-complexity DNA libraries through optimized design, cloning, barcoding, transformation, and NGS QC workflows that reduce bias, dropouts, frameshifts, chimeras, and sequencing artifacts. We use multi-agent AI workflows to accelerate library design, identify manufacturability risks, and interpret NGS data. However, AI cannot rescue flawed inputs: better libraries produce better data, and better data produces better models.