From molecular evidence to biochemical understanding.

Scope: computational questions with biological context

Biology is structured across scales: sequence, molecular structure, interaction, pathway, cell, tissue, organism, and population. A computational result is interpretable only when it identifies the scale and system it represents. Flintglade's direction emphasizes tools and methods that keep those distinctions visible while organizing public research, comparing computational approaches, or generating hypotheses for later testing.

The current public work includes this defined research scope and Oxalosuccinate — Molecular Pathway Evidence. The platform applies the same evidence practice to a connected human-metabolism map and molecular records, then carries that context into biochemical learning surfaces.

A public biochemistry learning layer

The Flintglade Biochemistry Learning Library is built beside the research platform rather than separated from it. Readers can trace human metabolism, work through all 22 native biochemical-mechanism families, practice with 323 topic-filtered biochemistry flashcards, and download nine complete guides for free. The wider designed-PDF collection contains 18 biochemistry guides organized across foundations, metabolism systems, and mechanism reasoning, while one Calculus II guide remains separate.

The guide language and visual presentation are structured for a broad audience while preserving the original scientific reasoning, nuance, and edge cases. Material on amino acids, proteins, enzymes, kinetics, energetics, biomolecules, and metabolism can assist MCAT biochemistry study while remaining grounded in the wider subject rather than a single test format. The descriptive library catalog connects each guide to the free map, mechanism, and recall layers.

Research questions

How should molecular records preserve provenance?

A protein identifier, sequence, structure, assay value, chemical representation, or disease annotation can change meaning across database releases and experimental contexts. Flintglade is interested in evidence records that retain the originating database or paper, accession or compound identifier, organism or construct, experimental method, units, conditions, release or publication date, and transformations applied after retrieval.

What can machine learning infer from sequence or structure?

Learned representations can support classification, ranking, structure-related tasks, or property prediction. Their evaluation must account for homology, scaffold overlap, temporal leakage, repeated measurements, assay variation, and the distance between the training population and intended use. A random split that places close biological or chemical relatives on both sides can measure memorization-like transfer rather than useful generalization.

How should computational binding or design results be interpreted?

Docking, structure prediction, affinity estimation, generative design, and molecular simulation operate under different assumptions. A score can prioritize candidates within a defined workflow; it is not itself proof of binding, function, selectivity, exposure, efficacy, or safety. The useful question is whether the computation improves a stated decision compared with an appropriate baseline and whether its uncertainty is calibrated for that candidate space.

Can literature synthesis remain traceable at scale?

Life-science literature contains conflicting experiments, changing nomenclature, context-dependent effects, and multiple evidence types. A research system should link each extracted claim to the exact source, distinguish review articles from original experiments, preserve species and model-system context, and show when an inference connects facts that no source directly states. AI can assist discovery and organization, but it does not become a biological source.

Method: a molecular evidence record

A Flintglade computational study in this direction should define its target and validation ladder before choosing a model. Depending on the question, the evidence record includes:

• organism, tissue, cell line, construct, molecular entity, endpoint, and experimental context;
• source database or publication, accession or stable identifier, version, license or use terms, and retrieval date;
• the original units, assay protocol, censoring or detection limits, normalization, and transformations;
• dataset inclusion and exclusion rules, duplicate handling, relatedness-aware or time-aware splits, and leakage checks;
• the computational method, software and model version, parameters, random seeds, hardware context, and failure states;
• baselines that match the question, uncertainty estimates, calibration, subgroup or regime analysis, and negative results; and
• the next evidence step required: computational replication, orthogonal dataset, biochemical assay, cellular experiment, organism study, or clinical research.

Literature and database claims follow the public-source research method. Any AI model or benchmark claim also follows the AI and machine-learning methodology.

A validation ladder, not a collapsed score

Flintglade separates evidence levels because each answers a different question. A database annotation reports an attributed record. An in-silico prediction estimates an outcome under a model. Reproduction on a held-out dataset tests computational generalization. A biochemical assay observes a molecular effect under defined conditions. A cellular or organism study adds biological context. Clinical research addresses human outcomes under a governed protocol.

Movement up this ladder is not automatic. A strong retrospective metric may fail prospectively. Binding may not imply functional effect. Activity in an isolated system may not translate across a cell membrane or survive metabolism. An association may not identify a mechanism. Result pages should name the highest evidence level actually reached and the next step still required.

Current public artifact: Oxalosuccinate

Oxalosuccinate is a free, source-linked research and learning surface for inspecting biochemical pathway topology and molecular evidence without erasing experimental context. It brings a connected map of human metabolism together with protein-family dossiers, metabolite records, cofactor context, learning chapters, mechanism reasoning, and active recall.

The metabolic network is organized across biomolecules, catabolism, biosynthesis, energy transfer, cellular compartments, and shared intermediates. Evidence-bearing records keep their source context attached, while the learning tools create multiple ways to approach the same biochemical system: pathway-scale tracing, reaction-level reasoning, explanatory reading, and flashcard practice.

Artifacts and reconstruction

Suitable public artifacts may include a versioned source ledger, machine-readable dataset card, documented preprocessing pipeline, reproducible notebook, model or workflow card, evaluation splits, result table, error analysis, and limitations statement. Where biological data or software licenses restrict redistribution, the artifact should provide stable identifiers and a reconstruction procedure rather than republishing material without permission.

Oxalosuccinate is developed as a versioned research platform: source context remains attached to each evidence-bearing record, release identifiers make the underlying corpus traceable, and the broader research practice keeps established findings, computational experiments, hypotheses, and long-horizon questions clearly organized.

Connection to AI and physical science

The AI and machine-learning direction provides evaluation and system-design methods for learned biological models. The physics and quantum-computing direction shares questions about simulation, uncertainty, inverse problems, and the difference between a model and a measurement. The life-science context adds its own requirements: biological relatedness, assay conditions, scale, and a validation ladder that reaches beyond computation.

Published 13 July 2026 · Last materially reviewed 14 July 2026 · Research questions and corrections: support@flintglade.com