B. Takulapalli, C. Agu, R. Cook, W. Martelly, L. Gusghari
INanoBio Inc.,
United States
Keywords: drug discovery, monoclonal antibodies, artificial intelligence, high-throughput screening, biological validation, antibody kinetics affinity, lead candidate selection
Summary:
Therapeutic drug discovery is an extensive market in biopharmaceuticals, suffering from a 90% failure rate in late-stage development. Artificial intelligence (AI) has been utilized to assist in high-throughput therapeutic monoclonal antibody (mAb) design to improve target binding. AI methods can generate tens of thousands of candidates for evaluation, necessitating high-throughput screening methods with improved characterization data for lead discovery. In silico methods can down-select candidates for downstream biological validation, but can suffer from high false-positive rates for off-target binding predictions and validation is limited by the small number of biological assays available. Sensor-integrated proteome on chip (SPOC) was developed as a kinetic proteomics platform to address these high-throughput target validation and lead candidate selection challenges. SPOC utilizes plasmid DNA arrays to synthesize full-length proteins in vitro which are simultaneously capture-purified directly onto gold-coated surface plasmon resonance (SPR) biosensor chips with up to 2,400 pathogen proteins per chip. This production method democratizes kinetic proteomics by effectively reducing costs 10-100x compared to traditional recombinant protein workflows. Use of SPR real-time biosensing enables SPOC to be the only proteomics platform capable of simultaneously providing quantitative, qualitative, and kinetic (affinity) data in a single assay, at scale. In this study, SPOC Proteomics collaborated with an AI mAb design company as a proof-of-concept to validate SPOC as a mAb characterization tool. First, B cells were isolated from vaccinated or recovered patients and sorted for COVID Spike-specificity. B cell V(D)J regions were sequenced and 798 clonotypes were identified. Candidates were down-selected based on in silico affinity predictions and 16 final candidates were selected, of which 12 were sent to SPOC for biological validation of RBD binding and kinetic affinity measurements. The 12 AI-designed mAbs were initially screened on a SPOC chip containing 10 COVID RBD variants ranging from Wuhan to XBB.1.16. The mAbs had distinct variant binding profiles, particularly affinity changed for variants occurring after Omicron, where some mAbs lost binding and others had improved binding. A few mAbs bound all variants, before and after the Omicron wave. Results indicate good general correlation between SPOC and BioLayer Interferometry kinetic measurements. SPOC qualitative binding profiles were compared and correlated well to flow cytometry and fluorescent assays. The 12 mAbs were subsequently screened on a prototype pan-respiratory pathogen SPOC chip containing 210 unique proteins from 12 respiratory pathogens plus their variants to look for off-target binding or cross-reactivity for the AI-designed mAbs which were specifically targeted to COVID RBD domains. Studies are in progress to correlate the SPOC binding kinetics to neutralization results as a potential future mechanism to replace time consuming, cumbersome in vitro neutralization assays.