InSilico-K: a drug-agnostic digital twin showing that molecular weight, not molecular target, governs biologic delivery and its telmisartan-mediated rescue in non-small cell lung cancer

Technical Summary: InSilico-K and the Molecular Weight Governance of Biologic Delivery

Problem Statement
Modern non-small cell lung cancer (NSCLC) therapy is increasingly dominated by immunoglobulin-scale biologics, including bispecific antibodies, antibody–drug conjugates (ADCs), checkpoint inhibitors, and anti-angiogenic antibodies. These agents cluster within a narrow molecular weight range of approximately 145–153 kDa, roughly 300-fold heavier than the small-molecule tyrosine kinase inhibitors (TKIs) they often accompany. The paper posits that elevated tumor interstitial fluid pressure (IFP) acts as a convective barrier that throttles the delivery of this entire "weight class" of macromolecules into the tumor core. While the barrier is known to restrict macromolecules, the hypothesis tested here is that this restriction is a property of molecular weight rather than specific molecular targets, and that a single, safe antihypertensive agent, telmisartan, could universally decompress this barrier to enhance delivery for all IgG-scale biologics.

Methodology: The InSilico-K Digital Twin
The authors developed InSilico-K, an open, deterministic, drug-agnostic digital twin engine. The core innovation of this framework is that molecular weight (Da) is the only drug-specific input; new agents are added as data points rather than requiring code modifications.

The engine operates through a three-layer logic:

1. IFP Estimation and Decompression: Baseline IFP is estimated non-invasively using a triad of inputs: CT-derived tumor volume, mean arterial pressure (MAP), and serum albumin. The model simulates the effect of telmisartan as a first-order decay process, reducing IFP over a 7-day pre-operative window.
2. Effective Tissue Diffusion: This combines an IFP penalty (applied via a binary flag only to macromolecules) with steric exclusion based on molecular weight. Small molecules are modeled as unaffected by the IFP penalty.
3. Response Mapping (InSilico-pCR): Cumulative receptor occupancy is calculated by integrating target affinity, effective diffusion, albumin-dependent free fraction, and free drug concentration. The time-integral of occupancy is mapped to a probability of pathological complete response, termed InSilico-pCR. This metric is defined as a model output for comparative analysis (standard care vs. telmisartan) rather than a direct prediction of wet-lab outcomes.

The study simulated 17 approved or late-stage agents (nine macromolecular, eight small-molecule TKIs) for a representative virtual patient (tumor volume 8.0 cm³, MAP 100 mmHg, albumin 3.8 g/dL).

Key Results

• Bimodal Molecular Weight Landscape: The analysis of current thoracic oncology agents revealed a "molecular-weight cliff." Agents are bimodally distributed: nine macromolecules at ~145,000–153,000 Da and eight TKIs at ~406–561 Da, separated by a ~315-fold gap with no agents occupying the intermediate range.
• IFP Dynamics: In the virtual patient, baseline IFP was 24.0 mmHg. Under modeled telmisartan treatment, IFP decayed exponentially to 7.8 mmHg by Day 7, crossing the capillary filtration threshold.
• Class-Independent Transport Rescue: The reduction in IFP resulted in superimposable effective-diffusion trajectories for three distinct macromolecular classes (amivantamab, datopotamab deruxtecan, and pembrolizumab). All three showed an approximate 5.9-fold acceleration in transport. Conversely, the small molecule osimertinib showed no change in diffusion, confirming the model's premise that IFP penalties apply specifically to macromolecules.
• Convergent Response Gain: The modeled InSilico-pCR for macromolecular agents increased from a baseline of ~31–36% under standard care to ~89–93% with telmisartan. The reference agent (amivantamab) shifted from 34.2% to 91.5%. Small molecules derived no benefit from the intervention.

Significance and Claims
The paper argues that IFP decompression behaves as a property of a molecular-weight class rather than a specific drug-target interaction. Because the principle is tied to molecular weight, the authors claim the strategy does not expire when first-line therapies change.

The study proposes two primary implications:

1. Cardiology–Oncology Co-prescribing: Telmisartan is framed not as a new cancer drug, but as a "bridge" to improve the intratumoral delivery of sophisticated biologics. This suggests a new paradigm where cardiologists and oncologists co-prescribe to optimize drug delivery physics.
2. InSilico-Guided Minimal Resection (IGMR): If systemic therapy achieves computationally anticipated pathological clearance via IFP decompression, the authors propose repurposing surgery from a primary curative modality to a "minimalist adjunct." IGMR is defined as resection limited strictly to residual macroscopic disease, guided by the engine's per-patient eradication prediction.

Validation and Limitations
The authors acknowledge that InSilico-K is a hypothesis-generating model with stylized coefficients calibrated for physiological plausibility rather than fitted to patient-level data. It assumes isotropic diffusion and omits spatial heterogeneity. To address this, the paper outlines a minimal, falsifiable window-of-opportunity (Phase 0) study. This proposed trial would involve 12–20 patients receiving a single neoadjuvant dose of a macromolecular agent followed by 7–14 days of telmisartan. The primary endpoint would be the intratumoral drug-concentration gradient (core vs. periphery) measured via LC-MS/MS or mass spectrometry imaging, aiming to convert the framework from hypothesis to evidence.