AI-Driven Precision Oncology Startup aiPTO Unveils #1-Ranked DELPHAI Model, Bridging the "Scale Gap" From Virtual Cells to Virtual Tumors

BASEL, Switzerland--(BUSINESS WIRE)--aiPTO TechBio today released a preprint [link] describing DELPHAI (Deep Explainable Perturbation Heterogeneity-Aware Inference), its virtual tumor model for perturbation prediction. In two independent benchmarking frameworks (Wei et al., Nature Methods 2026; Radig et al., Genome Biology 2026), DELPHAI achieved the #1 ranking on differentially expressed gene (DEG) recovery, outperforming state-of-the-art virtual cell models. It predicts drug responses for previously unseen, out-of-distribution patients, with biologically explainable outputs.

Executive Summary

• #1-Ranked AI Model: First on DEG recovery in two independent benchmarks, and first overall in scPerturBench's out-of-distribution setting.
• Fitness Gating Architecture: Natively models cell-type-specific survival, addressing a key limitation of existing models, which assume all cells survive treatment.
• Living + Virtual Tumor Moat: Coupled to aiPTO's proprietary Individualized Patient Tumor Organoid (IPTO) platform (500+ patients across 40+ CNS tumor subtypes; Peng et al., Cell Stem Cell 2025).
• Value Thesis: A predictive preclinical platform that targets a leading cause of clinical attrition, tumor heterogeneity and resistance, to de-risk and accelerate pharma R&D.

The Industry Challenge: The "Scale Gap" in Virtual Cell Modeling

The convergence of AI and biomedicine has ignited a surge in virtual cell modeling. However, the field faces a fundamental scale gap: state-of-the-art models predict how isolated single cells behave, rely heavily on cell-line data, and implicitly assume that every cell survives treatment.

Real human tumors are not single cells or uniform cell lines; they are complex, heterogeneous ecosystems. Under therapy, drugs deplete sensitive populations while sparing resistant clones that ultimately drive relapse. A model that cannot predict cell survival cannot predict how a patient's tumor responds.

The Technology Solution: Bridging to Virtual Tumors via Fitness Gating

DELPHAI addresses this bottleneck by integrating a novel "Fitness Gating" mechanism into its neural architecture. By learning cell-specific survival directly from data, DELPHAI identifies which cell populations a drug depletes, predicting shifts in tumor composition alongside single-cell transcriptomic responses.

Having validated the architecture on public benchmarks, aiPTO is now training DELPHAI on single-cell drug-perturbation data generated by its proprietary IPTO platform, which retains each patient's native tumor microenvironment. The result is a closed-loop living+virtual tumor platform: the living model generates patient-tumor data, the virtual model learns from it, and its predictions are tested back in the living model. This moves the field from virtual cells trained on cell lines to virtual tumors trained on real patient tissue.

Investment Perspective: A Living + Virtual Tumor Platform De-Risking Precision Oncology

aiPTO is not building isolated algorithms; it is building a drug-response AI model for precision oncology grounded in proprietary patient-tumor data. The platform is designed to predict tumor-microenvironment shifts, target engagement, resistance, and counter-resistance opportunities before clinical trials begin. By targeting a leading cause of oncology trial failure, unmapped tumor heterogeneity and resistance, it offers pharma partners a scalable, defensible way to de-risk drug discovery.

About aiPTO

aiPTO TechBio, a spin-out of the German Cancer Research Center (DKFZ), develops a living+virtual tumor platform and first-in-class therapeutics for precision oncology, with a focus on brain cancer. By coupling its proprietary Individualized Patient Tumor Organoids (IPTO), the living tumor model, with DELPHAI, its virtual tumor model, aiPTO works to make patient drug response predictable. Together with pharma and academic partners, the startup aims to change how oncology drugs are discovered and matched to patients.

More information: www.linkedin.com/company/aipto/ | www.aiptobio.com