Partex Accelerates Antibody Engineering for Neurodegenerative Diseases Using NVIDIA-Accelerated AI / ML Platform

Paris, June 2025

Partex has reduced the end-to-end timeline for antibody optimization from > 700 hours to 48 hours – a reduction by 93%. This enables faster, cost-effective development of therapeutic antibodies for neurodegenerative disease targets, leveraging cutting-edge GPU-accelerated structure prediction and molecular dynamics simulations.

An In-house developed AI/ML antibody design solution accelerated by NVIDIA Hopper GPUs made this breakthrough possible.

Antibody Designing against antigen associated with neurodegenerative disorders

Partex improved the binding affinity and specificity of Light Chain Bioscience’s antibody sequences targeting an antigen associated with neurodegeneration.

Light Chain Bioscience, (LCB) is a Swiss Biotechnology company with a focus to bring native bispecific therapeutics to patients with unmet medical needs like neurodegenerative and infectious diseases.

Partex collaborated with protein engineering and neuroscience experts from LCB to build end to end AI workflow for traditional lead generation and selection process. The input dataset included light and heavy chain sequences, binding affinity values, and target characteristics, derived using phage display techniques.

"Our joint efforts to develop an end to end AI workflow enhanced the optimization process and increased diversity of sequences found to bind the targets of interest," said Jean-Philippe Courade, CSO, discoveric bio alpha Ltd.

AI-Accelerated Antibody Optimization Pipeline

Partex implemented a two-tiered pipeline integrating AI-guided sequence generation and GPU-accelerated structural modeling, built on NVIDIA.

Sequence Engineering

• Input sequences underwent quality control, affinity normalization, and IMGT V-QUEST translation.
• Advanced models such as AutoRegressive Convolutional Neural Network (ARCNN) and Markov chains generated site-mutated variants guided by LCB’s lead optimization principles.
• Novel sequences were scored using a combination of:
  • Analytic Hierarchy Process (AHP)
  • Amino Acid Preference
  • Sequence Classification Models
  • ProtBERT Regression Scoring
• Final candidates were shortlisted considering optimal AI scores and diversity-preserving clustering algorithms.

Structural Modeling

• Antibody 3D structures were modeled using the NVIDIA AlphaFold2 Multimer NIM, resulting in a 98% time reduction.
• MegaDock4.0 on Hopper GPUs enabled docking of 200 antibody-antigen complexes in under an hour—a 99% improvement.
• PRODIGY (PROtein binDIng enerGY prediction and Arpeggio (interatomic interactions mapping) were integrated for high-confidence pose selection.

MD simulation-based Affinity

• Binding free energy calculations were conducted using the NVIDIA NGC GROMACS container, scaling across 200+ sequences with Hopper GPUs.
• The estimated relative binding free energy helps in precise ranking of the top selected therapeutic candidates in just 7 hrs.

Impact Snapshot – Process Time Comparison

Outcomes Delivered

The revamped antibody engineering pipeline is now capable of completing a full optimization cycle in just 2 days, compared to the traditional multi-week process. Key outcomes:

• 93% time reduction across pipeline stages
• $1,000 USD per-project cost savings by replacing licensed CPU-based tools
• Minimal manual intervention, allowing near-complete automation
• Enhanced accuracy of binding free energy predictions using MD simulation

"With this GPU-accelerated pipeline, we are enabling the next generation of biologic design—delivering quality, speed, and reproducibility at a scale previously impossible," said Dr. Gunjan Bhardwaj, CEO & Co-Founder, Partex.

Why This Matters

Partex’s AI/ML antibody design solution, powered by NVIDIA Hopper GPUs, AlphaFold2 Multimer, and GROMACS containers, represents a breakthrough in AI-driven drug discovery and Agentic AI. The platform allows researchers to:

• Predict accurate antibody-antigen interactions
• Validate hits using thermodynamic simulations
• Scale high-throughput design without compromising on details