← Back to Portfolio
Biomedical AI × Drug Discovery

DIPG Therapeutic Discovery

Artificial Intelligence-Driven Design of Novel Brain-Penetrant Therapeutics for Diffuse Intrinsic Pontine Glioma — A Lethal Pediatric Cancer

<1% 5-Year Survival
AI De Novo Design
>80% BBB Penetration
12mo Development

The Urgency: A Cancer With No Survivors

Diffuse Intrinsic Pontine Glioma (DIPG) is a rare, inoperable brainstem tumor that kills children within one year of diagnosis. Despite over 200 clinical trials, no chemotherapy has succeeded. The 5-year survival rate is under 1%. This demands revolutionary, out-of-the-box innovation.

The Paradigm Shift: AI-Designed Therapeutics

Traditional drug discovery relies on trial-and-error laboratory screening of thousands of compounds — a process taking years and often yielding nothing. Our approach harnesses artificial intelligence to design completely novel therapeutic molecules optimized for DIPG's unique vulnerabilities.

Instead of brute-force robotic screening, we deploy generative AI models and machine learning to extrapolate from vast biomedical datasets and chemistry libraries, suggesting drug candidates that human researchers would never consider. This is intelligence-guided exploration of chemical space at unprecedented scale.

Multi-Objective Optimization Target:

$$f(molecule) = w_1 \cdot \text{Potency}(IC_{50}) + w_2 \cdot \text{BBB}_{\text{penetration}} + w_3 \cdot (1 - \text{Toxicity})$$

AI simultaneously optimizes across competing objectives: nanomolar inhibition of target (IC₅₀ < 1μM), blood-brain barrier permeability (>80% penetration), and minimal off-target toxicity. Traditional medicinal chemistry tackles these sequentially; AI explores the entire parameter space in silico.

The Science: Targeting DIPG's Achilles' Heel

Metabolic Vulnerability Discovery

Recent peer-reviewed research reveals that DIPG tumors harbor distinctive metabolic dependencies absent in normal pediatric brain tissue. The hallmark H3K27M histone mutation creates unique epigenetic and biosynthetic requirements that can be therapeutically exploited.

Target Identification

  • DHODH enzyme (pyrimidine biosynthesis)
  • NAD⁺ metabolism pathways
  • De novo nucleotide synthesis addiction
  • Epigenetic reader domain proteins
  • H3K27M-specific vulnerabilities

Validation Strategy

  • siRNA/CRISPR knockdown in DIPG cell lines
  • Patient-derived xenograft models
  • 3D tumor organoid viability assays
  • Genomic data mining (TARGET, CBTTC)
  • Selectivity vs. normal astrocytes

Blood-Brain Barrier Challenge

  • Brainstem location mandates BBB penetration
  • Traditional CNS drug failure: <2% BBB crossing
  • AI prediction of passive permeability
  • Active transport mechanism design
  • Nanoparticle delivery alternatives

BBB Permeability Prediction Model:

AI models learn from thousands of compounds with known BBB penetration, encoding molecular features as multidimensional vectors:

$$P_{BBB} = \sigma \left( \sum_{i=1}^{n} w_i \cdot \phi_i(molecule) + b \right)$$

Where φᵢ are learned molecular descriptors (lipophilicity, hydrogen bonding, polar surface area, molecular weight), wᵢ are trained weights, and σ is a sigmoid activation yielding probability of CNS penetration. Molecules scoring P_BBB > 0.8 advance to synthesis.

The AI Drug Design Pipeline

Phase I Research & Development Workflow

Phase 1
Target Validation

Confirm chosen biological target drives DIPG vulnerability via in vitro genetic knockdown and pharmacological inhibition. Establish IC₅₀ threshold for efficacy.

Phase 2
AI-Driven Lead Generation

Deploy generative models (VAE, GAN, transformer-based) to design novel small molecules. Rank candidates by predicted target affinity, BBB penetration, and synthetic accessibility.

Phase 3
Medicinal Chemistry Synthesis

Synthesize top 5-10 AI-designed compounds. Characterize chemical stability, solubility, and structural confirmation via NMR/MS.

Phase 4
In Vitro Efficacy Testing

Test compounds on DIPG patient-derived cell lines and 3D organoids. Measure IC₅₀, selectivity index vs. normal cells, and mechanism of action validation.

Phase 5
Pharmacokinetics & In Vivo Proof-of-Concept

Mouse PK study: measure plasma/brain tissue drug levels via LC-MS. Pilot efficacy in orthotopic DIPG xenograft: tumor volume reduction, survival extension.

Phase 6
Iterative Refinement & Reporting

Feed experimental results back to AI for model retraining. Optimize lead compound for Phase II development. Compile proof-of-concept data for NCI SBIR submission.

Generative Model Architecture:

Variational Autoencoder (VAE) or Generative Adversarial Network (GAN) learns latent representation of "drug-like" chemical space:

$$z \sim \mathcal{N}(0, I), \quad molecule = G(z; \theta_G)$$

Generator G decodes random latent vector z into valid molecular SMILES strings. Discriminator or property predictor scores outputs for target inhibition and BBB permeability. Gradient ascent optimizes z to maximize desired properties, generating novel molecules outside training distribution.

Why This Is Genuinely Transformative

Novel Target & Modality

Not repurposing existing drugs — designing first-in-class molecules against DIPG-specific metabolic vulnerabilities. Opens entirely new therapeutic avenue where 200+ previous trials failed.

AI-Powered Discovery Paradigm

Unprecedented application of generative AI to pediatric oncology. Reduces discovery timeline from 5-7 years to 12-18 months. Demonstrates scalable platform for other rare cancers.

High Risk, High Reward

Significant risk: predicted target may not translate to patients, chemistry may fail in vivo. Commensurate reward: first effective DIPG therapy would save hundreds of children annually and revolutionize pediatric neuro-oncology.

NCI SBIR Alignment

Directly addresses Innovative Concept Award criteria: transformative (not incremental), novel technology (AI de novo design), high unmet need (DIPG mortality), small business commercialization pathway.

Broader Impact & Platform Potential

Success creates replicable methodology for other pediatric and rare cancers currently neglected due to small patient populations. The AI-drug platform becomes commercializable technology licensing to pharmaceutical partners for multiple indications.

Expected Milestones & Success Criteria

Phase I Deliverables (12 months)

  • Validated DIPG metabolic target with genetic/pharmacological proof
  • 5-10 novel AI-designed molecules synthesized
  • At least 1 compound with IC₅₀ < 1μM in DIPG cells
  • Demonstrated selectivity: >10x window vs. normal astrocytes
  • PK data confirming brain penetration in mice
  • Pilot in vivo efficacy: tumor growth delay or survival benefit

Phase II Goals (24 months)

  • Lead optimization: improve potency, PK, safety
  • IND-enabling toxicology studies (GLP)
  • Manufacturing scale-up for clinical supply
  • Regulatory pre-IND meeting with FDA
  • Partnership with pediatric oncology center
  • IND submission for Phase I clinical trial

Long-Term Vision

  • First-in-human Phase I trial in relapsed DIPG patients
  • Biomarker development for patient selection
  • Combination therapy strategies (radio-sensitization)
  • Expand to other pediatric brain tumors
  • Establish AI drug design as standard in rare cancers
  • Change survival curve: from 0% to measurable long-term survivors

Partner in Pediatric Cancer Innovation

This project represents the fusion of cutting-edge AI with urgent medical need. If you're a funding agency, pharmaceutical partner, or research institution ready to make history in pediatric oncology, let's collaborate.

Discuss DIPG Project

← Back to Portfolio