Building the AI-Ready Foundation for Antibody Discovery

Every AI model applied to antibody discovery is only as good as the data underneath it. From developability prediction to affinity ranking to lead selection, predictive models require a robust, structured, reproducible data foundation, and most discovery pipelines aren't built to produce one. In this webinar, the MiLaboratories team shows how Platforma — antibody discovery platform built on the open-source MiXCR toolkit — takes you from raw NGS data through clonotyping, clustering, enrichment analysis, developability assessment, and lead selection in one reproducible workflow. It's aimed at wet-lab scientists, computational biologists, and discovery leaders who want practical approaches to prioritizing antibody leads and preparing their pipelines for machine learning integration. What you'll learn: -Repertoire characterization for AI: verify your panning campaign actually captures the high-quality, reproducible data predictive models require, using rarefaction and clonal-dominance checks. -Antibody clustering for machine learning: group redundant sequences to reveal unique binding mechanisms and structure your data for downstream ML — an alternative to stitching together MiXCR, custom scripts, and spreadsheets. -Decoding selection dynamics: analyze enrichment trajectories, binding specificity, and cross-antigen profiles to separate stable true binders from non-specific "parasites." -In silico de-risking: automatically flag sequence liabilities early to avoid synthesizing dead ends and keep your AI training data clean. -Building an AI-ready loop: map wet-lab assay results back to sequence lineages to optimize variants, standardize team workflows, and build a reusable foundation of training data. Methods and tools covered: rarefaction analysis for sequencing depth, Gini index for clonal dominance, MMseqs2 sequence clustering (95% identity), Parapred paratope clustering, ImmuneBuilder structure prediction and structure-based clustering, log2 fold-change enrichment analysis, enrichment-quality classification (stable binder, rescuer, weak binder, parasite), negative-control specificity filtering, UMAP clonotype-space visualization, sequence-liability and developability scoring, and a two-step panel selection funnel from ~500 candidates to ~96 validated leads. Includes a live demo on a published phage display study. Questions this session answers: What software takes you from FASTQ files to ranked antibody leads? What is the best software for phage display NGS analysis? How do you combine enrichment and developability when selecting antibody leads? What can you use alongside MiXCR for downstream antibody discovery analysis? How do you track antibody clonotype enrichment across panning rounds? Learn more: https://platforma.bio