Video summary

Accelerating Bioprocess Development with AI + Automation

Main summary

Key takeaways

Science and Nature

Scientific concepts, discoveries, and nature phenomena presented

  • AI for bioprocess development is constrained by data quality

    • AI models are only as useful as the experimental data used to train them.
    • Bioprocess data must be comparable across experiments (i.e., “apples to apples”).

  • Need for traceable, structured experimental datasets

    • Challenges arise because bioprocess datasets are often:
      • Unstructured or inconsistently formatted
      • Inconsistently annotated (metadata/context differs across scientists and sites)
      • Stored in multiple systems (spreadsheets, ELNs, LIMS, historians, even paper)

  • Operator-to-operator variability in bioprocessing

    • A key source of experimental “noise” is variation introduced during manual execution.
    • Reducing this variability increases signal-to-noise ratio, improving confidence and reducing time/cost to reach robust manufacturing conditions.

  • Metadata/ontology standardization (“AI-ready data”)

    • “AI-ready” implies datasets include:
      • Consistent ontology
      • Consistent metadata
      • Standardized format enabling comparison across:
        • scientists,
        • time points,
        • sites,
        • and sponsors/CDMOs.

  • Automated manufacturing/data generation for monoclonal antibody (mAb) purification

    • The talk centers on high-throughput monoclonal antibody production/purification workflows.
    • A specific automation platform is used to reduce variability and accelerate data generation.

Methodologies / approaches (outlined as presented)

  • Axio’s federated AI/data approach

    • Federated network (local nodes)
      • Install API/controllers (“nodes”) at pharma sponsors and CDMOs/vendors.
      • Data stays local on partner servers; Axio does not centralize the raw data to its servers.
      • Nodes create data pipelines that extract/pull data from existing systems (e.g., LIMS, ELNs, historians, and instruments).
    • Coordination and translation layer
      • Translate partner data into a shared single ontology so datasets become comparable across sites.
      • Focus initially on process development lab → GMP manufacturing lab handoff (where transcription/translation time and errors are common).
    • Why translation is necessary
      • Sponsors may try to impose a single ontology, but partners work with many sponsors, making “one ontology for everyone” impractical.
      • Therefore, a coordination/translation layer sits above existing systems rather than replacing them.

  • Hybrid path to automated translation

    • Current programs: manual approach for building/creating standardized translations into a uniform ontology.
    • Future programs: AI-enabled (“agentic”) automated extraction/translation during real-time or near-real-time data ingestion, using model-driven transformation into the target ontology.
    • Uses community/industry structure references (mentioned: Pistoia Alliance) as a basis for structuring data.

Case study / experimental findings described

  • Automation reduces variability in mAb purification

    • Comparison: traditional manual magnetic bead setup vs automation using Genens’s Quattro ProAB 1300 (automated magnetic bead platform).
    • Setup: 8 identical antibodies, same purification protocol; only downstream purification execution differs (automated vs hands-on).

  • Key measured metrics

    • Yield (improved modestly with automation)
    • Dispensing volume variation (dramatically reduced with automation)
    • Hands-on time (reduced from “a couple of hours” to a small fraction)

  • Quantified takeaway

    • Reported nearly 90% reduction in hands-on time.
    • The primary benefit emphasized is reduced operator-driven variability, improving comparability and lowering effort to reconcile differences later.

  • Scaling goals

    • Move from ~30 mL expression volume prior experiments (run in Falcon tubes) toward more automated formats:
      • Quattro Mini 1100 for 96-well
      • and 48-/24-well formats
    • Target: 20,000–30,000 data points per week for monoclonal antibody workflows.

Researchers or sources featured (named in subtitles)

  • Justin Buyers — Founder and CEO of Axio Biioarma/BioPharma (as stated in the subtitles)
  • Pistoia Alliance — referenced as an example of an organization with data-structuring guidance (mentioned in a question/answer)
  • Genens — mentioned in relation to the Quattro ProAB 1300 magnetic bead platform

Original video