DNAI Biotech Launches Physics-Constrained Digital Twin Platform for Personalized Oncology

STOCKHOLM / SAN FRANCISCO — February 2026 — DNAI Biotech today announced the launch of its physics-constrained digital twin platform for personalized oncology, alongside three provisional patent filings covering novel methods for translating preclinical AI models to human patients. The platform, live at www.dnai.bio , converts multi-omics patient data into continuous tumor evolution simulations — enabling clinicians and drug developers to predict treatment response, resistance timing, and optimal dosing for individual cancer patients.

Unlike conventional AI approaches in oncology that produce static binary classifications ("responder" vs. "non-responder"), DNAI generates dynamic, continuous-time trajectories of tumor evolution under different treatment scenarios, complete with calibrated uncertainty estimates. The platform supports 33 cancer types, with a primary focus on soft tissue sarcoma.

The Technology

DNAI's dual-path architecture combines two complementary paradigms:

• Interpretable analysis (V1): A modular pipeline that identifies driver genes (AUROC 0.954 on the IntOGen benchmark), ranks therapy options, and predicts survival risk using a graph attention network and domain-adapted drug response models.
• Dynamic simulation (V2): A neuro-symbolic simulator that encodes patient biology into the parameters of differential equations governing tumor growth, clonal competition, and drug response. This enables gradient-based treatment optimization — computationally searching for the drug combinations and dosing schedules that minimize tumor burden while respecting toxicity constraints.

The platform ingests six data modalities — RNA expression, copy number variation, DNA mutations, methylation, histology, and medical imaging — and handles missing modalities through a novel uncertainty-calibrated imputation method.

Three Provisional Patents

DNAI has filed three provisional patents addressing fundamental challenges in translating AI models from preclinical data to clinical use:

1. Physics-Constrained Sim-to-Real Transfer Learning (App. No. 63/967,576) — A domain separation network that corrects for biological artifacts when transferring models trained on patient-derived xenograft (PDX) mouse models to human tumors, including stroma replacement artifacts and data sparsity differences.
2. Preventing Metabolic Scaling-Induced Representational Collapse — A selective adversarial alignment method that preserves proliferation-rate signals during cross-species transfer by identifying and excluding metabolism-linked latent channels from domain adaptation, using allometric scaling correction.
3. Uncertainty-Calibrated Missing Modality Imputation with Multi-Layer Uncertainty Quantification — A three-layer uncertainty system combining parameter ensembles with population priors, time-dependent trust that reverts to population baselines when patient data is sparse, and oscillation detection for numerical stability.

Validation Results

Applications

For clinicians: Patient-specific therapy ranking, continuous tumor trajectory predictions with confidence intervals, and resistance emergence timing — transforming molecular tumor board discussions from static genomic reports into dynamic simulation-based treatment planning.

For biopharma: Virtual clinical trial simulation, gradient-based dose optimization, mechanism of action discovery, and patient stratification — reducing the cost and time of oncology drug development by enabling in silico experimentation before enrollment.

About DNAI Biotech

DNAI Biotech is a computational oncology company building physics-constrained AI for personalized cancer treatment. Founded by Per Magnus Swedenborg, the company's mission is to transform oncology from a statistical guessing game into a precise, predictive science by encoding the laws of physics into the heart of artificial intelligence.

The platform is currently designated for research use, with a regulatory pathway toward FDA Class II Software as Medical Device (SaMD) clearance via the de novo classification process.