Cracking the Immune System with AI

Episode Summary

Executive Summary: The episode features Immuni scientific founder Danny Wells explaining how the company uses AI and single-cell genomics to map the immune system and improve immunotherapy development. He argues that many checkpoint inhibitors work differently than originally believed, and that scalable, reproducible immune profiling can reveal mechanisms, biomarkers, resistance, and better drug combinations across discovery, translational, and clinical settings.

Main Topics: Immuni’s mission to map the immune system (Priority: 5/5): Wells describes Immuni as an AI-enabled, single-cell profiling platform aimed at building a granular map of immune cell types, interactions, and cytokine signaling to better understand health and disease. Reframing how PD-1 inhibitors work (Priority: 5/5): He cites work in basal cell carcinoma patients treated with pembrolizumab showing that anti-PD-1 therapy may recruit new tumor-specific T cells rather than merely reinvigorating resident ones. Need for scale and reproducibility in immune profiling (Priority: 5/5): Wells argues that answering immune-mechanism questions in humans requires longitudinal, high-dimensional data across many patients and time points, with strong controls for batch effects and sample handling. Role in translational and drug development workflows (Priority: 4/5): Immuni positions itself as a collaboration partner for biotech and pharma, helping characterize drug effects in humans, identify biomarkers, and inform combination strategies before and during development. From raw data to actionable insight (Priority: 4/5): The company emphasizes a flexible delivery model that can provide raw data, processed data, analyses, figures, and interpretive insights tailored to partner needs. Precision medicine beyond the tumor (Priority: 4/5): Wells says immunotherapy response prediction will require measuring the peripheral immune system, not just the tumor, to assess efficacy, toxicity, resistance, and patient-specific variability.

Key Arguments: Single-cell genomics is necessary to understand the immune system at the level where drug mechanisms and patient heterogeneity become visible. The anti-PD-1 mechanism may involve de novo recruitment of tumor-specific T cells from the periphery, challenging the long-held reinvigoration model. Academic labs can generate proofs of concept, but company-scale infrastructure is needed to process hundreds of samples reproducibly across trials. Drug developers need longitudinal human immune data to understand variability, biomarkers, and whether a therapy is acting as intended. Mapping the immune system creates a foundation for computationally testing perturbations and designing better therapies and combinations. Reproducibility is a major challenge in genomics, so Immuni vertically integrates lab processing and machine learning normalization to reduce batch effects. Precision medicine for immunotherapy will require biomarkers from both the tumor and the systemic immune compartment. The end goal is not just data generation but an 'aha moment' that directly informs program decisions and next-step development.

Data Points: Patients in pembrolizumab/BCC study: ~10 patients - The founder described an early collaboration with Stanford on basal cell carcinoma patients treated with pembrolizumab. Time points per patient in study: 2 time points - Samples were collected before and after treatment in the described study. Total study samples: ~20 samples - Wells framed the initial dataset as roughly 10 patients with two samples each. Illustrative scale target: 20 to 200 samples - He said the company was motivated by the desire to scale from an academic pilot to a much larger clinical-trial workflow. Per-patient sampling cadence: 3 to 5 samples - He described typical longitudinal blood collection over time during a trial. Cell count per sample: Thousands to tens of thousands of cells - Used to illustrate the dimensionality of each assay run. Genes measured per cell: Thousands to tens of thousands of genes - Wells used this to explain the data burden of single-cell profiling. Total data volume: Billions to tens/hundreds of billions of measurements - He estimated the combined scale across genes, cells, samples, and patients. Data volume per vial of blood: ~1 terabyte - He referenced a commonly cited scale for the amount of data produced from a single vial of blood.

Pivotal Quotes: "what people have thought for a long time is that the mechanism of action of these drugs... is that it's going into the tumor and reinvigorating cells that are already there" — Danny Wells: Explaining how the Stanford collaboration challenged the prevailing model for anti-PD-1 therapy. "we're really trying to drive to the aha moment with the data that we generate in collaboration with our partners" — Danny Wells: Describing Immuni’s goal beyond raw data generation. "We have this huge amount of dark matter there is in the immune system and highlights the need to study that dark matter and help bring light to it" — Danny Wells: Summarizing why immune mapping remains an open scientific frontier.

Implications: The discussion suggests immunotherapy development will increasingly depend on scalable single-cell immune profiling, better reproducibility, and human longitudinal data. For drug developers, that could mean faster biomarker discovery, fewer trial misfires, and more rational combinations.