The Best Argument for AI in Biotech Isn't Speed — It's What It Could Do for Rare Diseases

Rare diseases have a structural economics problem that no amount of scientific goodwill has ever fully solved. Individually, each condition affects too few people to justify the standard economics of drug or diagnostic development. Collectively, rare diseases affect an estimated 30 million people in the EU alone, across roughly 6,000 to 8,000 identified conditions — most with no approved treatment and many with no reliable diagnostic pathway at all. The Orphan Drug Regulation (EC 141/2000) exists precisely because the market, left alone, systematically underinvests in exactly this population. Incentives like ten years of market exclusivity and reduced regulatory fees have helped, but they address the revenue side of the equation, not the cost side. And for most rare disease programmes, the cost side is where the real barrier sits.

That is the part where AI-driven efficiency is starting to make a genuine, measurable difference — not through some abstract notion of "AI will cure diseases," but through the unglamorous compression of specific, expensive steps in the pipeline that have always disproportionately burdened rare disease programmes relative to their patient population.

Where the cost actually concentrates in rare disease work

The diagnostic odyssey is itself a cost, borne mostly by patients and health systems, before a treatment programme even exists. The average time to diagnosis for a rare disease is still commonly cited at several years, often after multiple misdiagnoses and specialist referrals — not because clinicians are careless, but because a given clinician may see a specific rare condition once in a career, if ever. AI-assisted variant interpretation and phenotype-matching tools, trained across aggregated case data rather than one clinician's individual experience, are measurably shortening that odyssey in early deployments. A faster, more accurate diagnostic pathway does not just help the individual patient — it also generates the very case identification that a future clinical programme needs to even assess feasibility.

Natural history data is expensive to assemble and used to require years of dedicated registry-building. Regulatory bodies increasingly accept natural history studies and real-world data as part of the evidence package for rare disease programmes, recognising that a traditional randomised controlled trial is often infeasible when the entire global patient population numbers in the hundreds. Assembling that natural history evidence traditionally meant years of manual chart review and registry curation. AI-assisted extraction and structuring of clinical data from health records, case reports, and existing registries compresses that timeline meaningfully — turning what was a multi-year prerequisite into a task measured in months, without lowering the evidentiary bar, only the labour cost of meeting it.

Regulatory documentation for orphan designation and beyond is a fixed cost that lands disproportionately hard on small programmes. Preparing an orphan drug designation application under the EMA's framework, or the equivalent FDA pathway, requires a defined prevalence justification, a medical plausibility case, and a structured regulatory dossier — work that costs roughly the same in person-hours whether the eventual patient population is 50,000 or 500. For a well-funded programme targeting a common indication, that fixed cost is a rounding error. For a small biotech pursuing an ultra-rare indication, it can be a meaningful fraction of total programme cost. AI-assisted drafting, literature synthesis, and regulatory precedent research — used with the verification discipline that this kind of documentation demands — measurably reduces that fixed cost, which matters most exactly where the patient population is smallest and the economics were already thinnest.

Why this changes the calculus for small biotech specifically

The organisations best positioned to benefit from this are not the large pharma companies that already have dedicated rare disease divisions and can absorb fixed costs across a large portfolio. They are the small, often academically-spun-out biotech SMEs that have a genuine scientific insight into a specific rare condition but have historically had to make an early, painful call: pursue the rare indication they actually understand, or pivot toward a larger, more commercially conventional indication because the smaller one could not carry the fixed cost of development.

Lowering the marginal cost of diagnostic pathway development, natural history evidence generation, and regulatory documentation shifts that calculus. It does not make a rare disease programme cheap — clinical development, manufacturing, and the core science remain expensive regardless of how efficient the surrounding pipeline gets. But it changes which programmes clear the bar of "economically defensible to attempt" for a resource-constrained SME, and that bar has historically excluded a lot of scientifically sound rare disease work that simply could not justify itself on a spreadsheet.

The caveat that has to come with this

None of this works without the same rigor discussed elsewhere on this blog regarding AI in regulated life science contexts. Variant interpretation feeding a diagnosis, natural history data feeding a regulatory submission, and documentation feeding an orphan designation application are all, without exception, high-stakes outputs that require the same verification discipline as any other AI-assisted work in a regulated environment — a qualified human reviewing and attesting to the result, a documented process, and a clear line between AI-assisted drafting and validated conclusion. The efficiency gain is real, but it is an efficiency gain in the labour of assembling evidence, not a shortcut around the evidentiary bar itself. Treating it as the latter would undermine the exact credibility that rare disease programmes, more than most, depend on.

With that discipline in place, though, this is one of the genuinely optimistic stories in an AI adoption conversation that is otherwise, appropriately, full of caveats. A technology that lowers the fixed cost of pursuing evidence-based, rigorously validated treatment pathways is a technology that expands which diseases the system can afford to take seriously — and for the patients behind those 6,000 to 8,000 rare conditions, that shift in what is economically defensible is not an abstraction.

---

If your organisation is evaluating whether a rare disease programme is viable, or trying to make an existing one more efficient without compromising the evidentiary rigor it depends on, that is a conversation worth having early. Get in touch if that is where you are.