Streamline Antibody Developability with DeepViscosity Now Available on Vecura

What is DeepViscosity?

DeepViscosity is an ensemble deep-learning artificial neural network model designed to predict the high-concentration (150 mg/mL) viscosity class of monoclonal antibodies (mAbs). By analyzing heavy- and light-chain variable-region amino-acid sequences, it classifies antibodies as either "Low" (≤ 20 cP) or "High" (> 20 cP) viscosity. It helps users streamline the drug discovery process by identifying high-viscosity candidates early, avoiding the need for expensive and time-consuming physical formulation screening for unsuitable candidates. It is especially useful for antibody engineering and therapeutic development teams focused on subcutaneous delivery.

What can users do with DeepViscosity on Vecura?

With DeepViscosity on Vecura, users can:

• Predict the viscosity classification of monoclonal antibody candidates directly from amino-acid sequences.
• Obtain an ensemble mean probability and confidence measure (standard deviation) to assist in candidate ranking.
• Generate intermediate DeepSP spatial-property descriptors (SAP and SCM scores) to gain insights into the biophysical drivers of viscosity.
• Efficiently triage therapeutic candidates before committing resources to wet-lab synthesis or formulation trials.

What the output means

The output provides a binary viscosity classification (Low/High), the mean class probability across a 102-member ensemble, and an ensemble standard deviation indicating confidence. Additionally, it returns a 30-column table of DeepSP spatial-property descriptors.

This output should be used to support scientific decision making. It does not replace experimental validation.

Why this matters

High-concentration viscosity is a critical developability property for therapeutic antibodies, particularly for subcutaneous administration. Previously, measuring this property required physical material, making it a significant bottleneck in the pipeline.

By leveraging DeepViscosity, researchers can predict potential manufacturing or delivery hurdles in silico. This early-stage insight allows teams to focus their resources on candidates with better biophysical profiles, significantly accelerating the path from discovery to clinical development.

• Developed by: Laila (Research Group)
• Source: Official GitHub Repository
• Reference: Original paper (mAbs, 2025)