Biochemistry

Podcast 2 — In conversation

Speaker A: I finished my biochemistry assignment last night — three thousand words on enzyme kinetics. My brain is completely saturated.

Speaker B: Enzyme kinetics! Now there is a topic. The Michaelis-Menten model?

Speaker A: The very same. I think I could recite the derivation in my sleep at this point.

Speaker B: Biochemistry has this beautiful quality of making you realise that everything you think of as biological is actually chemical at some fundamental level.

Speaker A: That is exactly what drew me to it. The moment you realise that a heartbeat is electrochemical, that memory is molecular — it reframes everything.

Speaker B: The word itself comes from bios — life — and chemeia, the root of chemistry. So literally: the chemistry of life. It is a very honest compound word.

Speaker A: And it covers an enormous range. Molecular biology, cell biology, genetics, metabolism, pharmacology — all have deep biochemistry roots.

Speaker B: Now in everyday English, biochemistry gets used in a slightly looser sense — the biochemistry of stress, the biochemistry of happiness. That is people referring to the hormonal and neurochemical processes rather than the academic discipline strictly.

Speaker A: Right — serotonin, dopamine, cortisol — the popular press loves talking about biochemistry when they mean specific neurotransmitters or hormones. It is a reasonable shorthand.

Speaker B: Though a biochemist would gently note that the field is rather broader than mood chemistry.

Speaker A: Ha — yes. A close related term is molecular biology, which overlaps considerably. The line between them is somewhat blurry — biochemistry tends to focus on the chemical reactions themselves; molecular biology focuses more on the information flow through DNA and RNA.

Speaker B: And biophysics is another neighbour — applying physics principles to biological systems. All three live on the same frontier.

Speaker A: One thing to be careful about: biochemistry is formal. You would not say I studied biochemistry at school — unless you went to a university. It is a degree-level or postgraduate discipline in Britain.

Speaker B: Though you might encounter elements of it in A-level chemistry or biology. The word tends to signal serious scientific engagement when used correctly.

Speaker A: It really does. Biochemistry is not a casual word — it carries the weight of the laboratory behind it.

Podcast 3 — Prompt engineering

Speaker B: Right — biochemistry is a rich domain for AI prompts because it involves structured scientific data: compounds, reactions, pathways, concentrations, assays. Every one of those is a distinct data modelling challenge.

Speaker B: Let us begin with a research laboratory information system.

Build a laboratory information management system for a biochemistry research lab. Track experiments with fields for experiment_id, researcher, start_date, status, and protocol_used. Each experiment has multiple sample entries with sample_id, compound_name, concentration in millimolar, pH, temperature, and observation_notes. Include a dashboard summarising active experiments by researcher.

Speaker B: Concentration in millimolar and pH are domain-specific units. Naming them produces the correct field types — decimal precision for concentration, a zero-to-fourteen range validator for pH — rather than generic numeric columns.

Speaker A: And protocol_used as a field is a detail that signals this is a regulatory-aware system. Labs need to track which standard operating procedure was followed.

Speaker B: Exactly right. Now a biochemistry education platform.

Create an interactive learning module for a biochemistry course platform. Build a metabolic pathway explorer where students can click on a pathway node — such as glycolysis, the citric acid cycle, or beta oxidation — and see the key enzymes, substrates, products, and ATP yield for that pathway. Display the pathway as a connected node diagram with highlighted active steps.

Speaker B: Three named biochemistry pathways give the AI real content to populate. It knows what glycolysis produces, what enzymes are involved, what the ATP yield is. Named domain knowledge in the prompt produces accurate scientific content without a data file.

Speaker A: The connected node diagram instruction is interesting — you are specifying a graph visualisation, not a table or a list. The rendering approach follows directly from the domain description.

Speaker B: Always. Now a pharmaceutical database.

Design a PostgreSQL schema for a drug-target interaction database used by a biochemistry research team. Tables: compounds with id, name, molecular_formula, molecular_weight, and SMILES string; targets with id, protein_name, gene_id, and organism; interactions with compound_id, target_id, binding_affinity_nM, interaction_type, and assay_method. Index on compound_id and target_id in the interactions table.

Speaker B: Binding affinity in nanomolar and SMILES string are real biochemistry data types. SMILES is the standard notation for molecular structure. Naming them produces a schema that a computational biochemist could actually use rather than a generic products table with a description column.

Speaker A: Assay_method is a really important field — it tells you how the binding affinity was measured, which affects how you interpret the number.

Speaker B: Precisely — provenance fields are critical in scientific databases. Now a health app.

Build a personal health tracking app that monitors biochemistry-relevant biomarkers. Users log blood test results including glucose in mmol per litre, cholesterol in mmol per litre, haemoglobin A1c as a percentage, and C-reactive protein in mg per litre. Display trend charts per biomarker over time and flag values outside the normal reference range with a colour-coded alert.

Speaker B: Four specific biomarkers with their clinical units removes all ambiguity. The AI generates correctly labelled axes, correct reference range logic — glucose normal is approximately three-point-nine to seven-point-one — and clinically appropriate alert colours.

Speaker A: Reference ranges are what make this medically meaningful rather than just a number logger. Did you hard-code them in the prompt?

Speaker B: I described the behaviour — flag values outside the normal reference range — and let the AI apply the clinical knowledge it has. For a prototype, that is sufficient. Production would use a validated medical data source. Now a bioinformatics pipeline dashboard.

Create a pipeline status dashboard for a bioinformatics team running biochemistry data analysis workflows. Each pipeline job has a name, status — queued, running, completed, failed — input file, start time, duration, and output file path. Display jobs in a sortable table. Failed jobs show a red alert with the error message. Completed jobs display a download link for the output file.

Speaker B: Four status states with specific UI treatments for each — queued is neutral, running is active, completed has a download action, failed has an error display — is a complete state machine specification in plain English. The AI generates the correct conditional rendering for every state.

Speaker A: That is four distinct UI components described implicitly through the state definitions. I would not have thought of it as a state machine, but that is exactly what it is.

Speaker B: Good. Now an enzyme kinetics calculator.

Build an enzyme kinetics calculator for biochemistry students. Accept inputs for substrate concentration S, maximum reaction rate Vmax, and Michaelis constant Km. Calculate reaction velocity V using the Michaelis-Menten equation: V equals Vmax times S divided by Km plus S. Plot a velocity-versus-substrate curve in a canvas chart. Show a Lineweaver-Burk double-reciprocal plot for the same data.

Speaker B: The Michaelis-Menten equation written in plain English is sufficient — the AI knows the formula and generates correct JavaScript. Lineweaver-Burk double-reciprocal plot is a specific biochemistry visualisation that the AI recognises by name and produces accurately.

Speaker A: Two named plot types, one equation, three input fields — and you have a scientifically valid teaching tool in one prompt. That is remarkable efficiency.

Speaker B: And a final one — a laboratory notebook system.

Build a digital laboratory notebook for a biochemistry research group. Each entry has a date, researcher name, experiment title, protocol reference, reagents used with quantities and units, observations, and conclusion. Entries are searchable by compound name, researcher, and date range. Include version history so supervisors can review and approve entries. Export selected entries as a formatted PDF report.

Speaker B: Reagents with quantities and units is a structured data spec that prevents free-text chaos — the AI generates a repeating form section rather than one large text area. Version history and supervisor approval describes a full review workflow in two words.

Speaker A: And PDF export for a regulated scientific environment is a genuine requirement — lab notebooks often need to be submitted to regulatory bodies or ethics committees.

Speaker B: Exactly. Domain knowledge in your prompt always produces domain-appropriate output. The word biochemistry signals to the AI: this is rigorous, data-heavy, scientifically precise work — and the output quality rises to match that expectation every time.