Xaira Therapeutics Unveils X-Cell Virtual Cell Model

In a recent episode of the Latent Space AI for Science podcast, Bo Wang, SVP and Head of Biomedical AI at Xaira Therapeutics, and Xi Chu, SVP of AI-enabled Discovery at Xaira, detailed the company's mission and the groundbreaking work behind their X-Cell virtual cell model.

Xaira's Mission in AI-Driven Drug Discovery

Xaira Therapeutics is at the forefront of leveraging artificial intelligence to accelerate drug discovery and advance patient care. The company is building a comprehensive AI platform that spans the entire drug discovery pipeline, from target identification and protein design to predicting patient responses to therapeutics.

Bo Wang explained that Xaira's unique approach lies in its AI-native, end-to-end strategy. "We aim to use AI to accelerate every part of the drug discovery," Wang stated, highlighting the goal of increasing success rates and significantly reducing the time it takes to bring new drugs to market.

The Three Pillars of Xaira's AI Platform

Xaira's platform is built upon three main AI capabilities:

• Protein Design: Leveraging advancements from Dr. David Baker's group at the University of Washington, Xaira aims to design novel molecules for previously undruggable targets.
• Virtual Cell / foundation model of biology: This platform, spearheaded by Wang and Chu, focuses on building AI models that can predict how cells respond to drugs and other interventions.
• Patient Representation Models: These models are designed to identify which patients are most likely to respond to specific therapeutics, crucial for precision medicine.

Chu emphasized the importance of integrating these three platforms to create a synergistic approach that makes drug development faster and more successful.

Introducing X-Cell: A Causal Virtual Cell Model

The conversation highlighted the recent release of X-Cell, Xaira's first virtual cell model. Xi Chu explained that virtual cells are AI models capable of predicting cellular functions and responses after specific interventions. Unlike purely descriptive models, X-Cell is designed to make causal predictions.

"For that, I think we need causal data," Chu stated, contrasting this with purely observational datasets. She elaborated on the concept of perturbations, where AI models can simulate the effect of altering a specific gene's expression within a cell to understand its broader implications.

The Power of Causal Data and High-Throughput Experimentation

The core challenge in building predictive biological models, according to Chu, is the availability of high-quality, causal data. Xaira is heavily invested in generating such data through high-throughput biology techniques, particularly perturb-seq.

This technique combines pooled CRISPR perturbations with single-cell RNA sequencing to create rich datasets. Chu detailed how CRISPR-Cas9 is used to disrupt gene expression, and how single-cell RNA sequencing allows for reading out the impact on all 20,000 genes simultaneously from each cell. This process generates a 2D dataset that powers the training of foundation models for biology.

Bo Wang noted the challenges in scaling these experiments, especially when handling hundreds of millions of cells. Xaira's engineering efforts have focused on industrializing the workflow, incorporating techniques like chemical fixation to lock cell states without compromising subsequent molecular biology steps, thereby ensuring high-quality data for their AI teams.

From Lab to Clinic: Bridging the Gap

The discussion also touched upon the critical transition from lab-based discoveries to clinical applications. While acknowledging the progress in protein design due to abundant data, Chu pointed out the data limitations in other areas like clinical model prediction and virtual cells.

Xaira's patient representation models are crucial for addressing this gap, aiming to predict patient responses to therapeutics. Wang expressed excitement about the integration of their three AI models, stating, "We always aim to connect three AI models together instead of letting them work individually by their own."

Ultimately, Xaira's mission is to transform drug discovery from an empirical process into an engineering discipline, making better drugs faster and with a higher success rate. The release of X-Cell marks a significant step in this direction, leveraging AI to unlock a deeper understanding of cellular biology and its response to interventions.