Materials & Industrial Discovery

In one sentence: The same engine that discovers drug candidates discovers new materials — swap the oracle from “drug-likeness” to “ionic conductivity” or “catalytic activity,” and the agents start building chemical intuition in a new domain.

Theory sources: SM (cumulative cultural evolution, Super-Ratchet), BM (meme dynamics, SIT-driven exploration), EMT (BLEND recombination, I-filter)

Implementation: bmc-engine/bmc-discovery/ — same codebase as drug discovery, different oracle and fragment library.


The Idea

BMC is a platform, not a point solution. In drug discovery, we proved the architecture works: QED 0.947, DRD2 0.997, cumulative knowledge across 40+ runs. The key insight: nothing in the engine is drug-specific. Agents explore fragment space, build hypotheses (meme graphs), test them against an oracle, and accumulate knowledge through cultural evolution.

Change the oracle and the fragment library — the cognitive machinery is identical.


Target Industries

Battery Electrolytes — $92B market, 18% CAGR

Problem: New battery chemistries (solid-state, sodium-ion) need new electrolytes. Current approach: DFT simulations (slow, expensive) or trial-and-error.

BMC approach: Oracle = DFT score (ionic conductivity + electrochemical stability). Fragments = molecular building blocks of electrolytes. Cultural evolution transfers knowledge: motifs that work in polymer electrolytes may work in solid-state.

Clients: CATL, Samsung SDI, LG Energy, Northvolt.

Industrial Catalysts — $41B market

Problem: Chemical industry depends on catalysts (oil refining, polymer production, green hydrogen). R&D is expensive and slow.

BMC approach: Oracle = activity/selectivity (computed or experimental). Fragments = ligands and metal centers. Promising case: CO$_2$ conversion catalysts (ESG trend, regulatory pressure).

Clients: BASF, Johnson Matthey, Haldor Topsoe.

Specialty Polymers — $85B market

Problem: Anti-corrosion coatings, biodegradable plastics, adhesives. Current approach: experiment + chemist’s intuition.

BMC approach: Oracle = target properties (tensile strength, glass transition temperature, biodegradability). Fragments = monomer blocks. CCE effect: knowledge from one polymer class accelerates the next.

Agrochemistry — $70B market

Problem: EU bans dozens of active substances yearly. Industry urgently needs new molecules. Average cost to bring one pesticide to market: $300M.

BMC approach: Multi-oracle optimization — molecule must be simultaneously active against pest, safe for bees, and soil-degradable. Three oracles in parallel. BMC already does this (cross-oracle pipeline proven in drug discovery).


Why BMC, Not ML

Neural networksBMC
Training dataRequired (often unavailable for novel materials)Not needed
HardwareGPU clusterStandard server
New material classCollect data, retrainPlug new oracle, run
ExplainabilityBlack boxFull trace: which agent found what, through which mechanism
Cross-domain transferCatastrophic forgettingCultural library persists across domains

What’s Already Proven (Drug Discovery)

The engine is validated. These are the drug discovery results using the identical architecture:

TargetBMC ScoreBest Specialized Algorithm% of Max
QED (drug-likeness)0.9470.94899.9%
GSK3$\beta$ (oncology)0.9601.00096.0%
JNK3 (neurodegeneration)0.7400.81091.4%
DRD2 (psychiatry)0.9970.99999.8%

Knowledge accumulates: fragment registry grew from 824 to 12,884+ across 40 runs.


What We’re Looking For

  • Domain partner with oracle access — a materials company with internal scoring functions (DFT models, experimental pipelines)
  • Pilot project — one material class, one target property, proof that cultural evolution transfers
  • Combined TAM of target industries exceeds $500B

For drug discovery details, see Drug Discovery. For the theory of cumulative cultural evolution, see Swarm Memetics.