Shivi Kumar
Biography
Research Interest
Abstract
Physics-Informed Neural Networks for Entropy-Constrained Collapse Dynamics and RNA?Small
Molecule Therapeutic Engineering in Hyperplastic Redundancy Networks of Pan-Cancer
Resistance
Drug resistance remains the dominant cause of cancer treatment failure, with tumors escaping
therapy through Hyperplastic Redundancy Networks (HRNs) that activate compensatory
survival loops. Current approaches develop ?second-line? drugs after resistance emerges, but
no existing strategy is designed to prevent resistance before it occurs.
We introduce CollapsePINN, an in silico framework that uses physics-informed neural networks
(PINNs) to model tumor dynamics as collapse systems. The model integrates mechanistic
priors?pharmacokinetics, dose-dependent resistance induction, hypoxia?tumor burden
coupling?with multi-omic and longitudinal clinical data to identify Synthetic Resistance Collapse
Points (SRCPs): minimal intervention sets whose disruption dismantles HRNs. CollapsePINN
outputs patient-specific collapse risk, predicted time-to-resistance, and optimized treatment
schedules to keep tumors outside the collapse basin.
Building on these predictions, we generate a new therapeutic class termed FusionComp
chimeras: dual-mode interventions pairing synthetic RNA strands that silence SRCP transcripts
with AI-curated small molecules that inhibit corresponding protein partners. These chimeras are
designed to be administered before or alongside standard therapy to collapse resistance
pathways preemptively.
In silico validation on semi-synthetic tumor growth ODEs and curated cancer cohorts (EGFRi,
BRAF/MEK, PI3K/AKT) demonstrates >40% improvement in early resistance detection and
~80% predicted gains in therapy durability compared to standard ML baselines. Current efforts
are extending toward wet-lab in vitro validation of RNA?molecule pairs in resistant cancer cell
lines, establishing the translational path from computation to experimental therapeutics.
This framework introduces a paradigm for resistance-proof oncology: a physics-grounded,
machine learning?driven platform capable of designing therapies that prevent drug resistance
across tumor types.
Keywords Cancer Resistance, Physics-Informed Neural Networks, Bioinformatics, Therapeutic Design,
Precision Oncology
