Computer-aided drug discovery has changed how research teams find promising molecules. Docking scores, molecular dynamics simulations, and binding-energy calculations can screen thousands of candidates in the time it once took to test a handful at the bench. For research institutions and biotech startups working with limited budgets, that efficiency is genuinely valuable.

But there’s a gap that trips up a surprising number of applicants: the same computational evidence that is powerful enough to guide discovery is often not sufficient to support a patent claim. If you are filing in the Philippines—and the principle travels well beyond it—a pharmaceutical or biotech application built on in silico data alone is vulnerable, and increasingly so.

The core problem: prediction is not demonstration

Philippine patent law requires that an application enable the invention. Under Rule 406.1 of the 2022 Revised IRR, the disclosure must contain “a clear and detailed description of at least one way of doing the invention using working examples,” leaving “nothing in conjecture.” For a claim whose novelty rests on what a compound does — inhibits a target, binds a protein, produces a therapeutic effect — that “doing” has to be shown, not forecast.

Computational modeling predicts an interaction. It does not demonstrate that the interaction actually occurs. An examiner looking at a docking study sees a hypothesis with numbers attached. What the enablement requirement asks for is evidence that the hypothesis holds—and a simulation, however sophisticated, is not that evidence.

This is why in silico results tend to collapse a whole cluster of requirements at once. When the only proof of activity is predicted, examiners commonly find the claims deficient not just for lack of enabling disclosure and full support, but also for industrial applicability (you cannot show the compound is useful in industry if you have not shown it works) and even inventive step (if a skilled person cannot reproduce the claimed technical effect from the disclosure, the effect cannot anchor an inventive-step argument). One weakness, several rejections.

The trap that cannot be undone later

Here is the part that catches applicants off guard, and it is the most important thing to understand: sufficiency of disclosure is assessed as of the filing date, and data generated afterward cannot cure it.

If you file on computational data and run your wet-lab assays six months later, that later data—however conclusive—generally cannot be added to the pending application to rescue it. The disclosure had to be sufficient when it was filed. This is not a formality that a persuasive response can talk its way around; it is a structural limit. An applicant who files too early, on prediction alone, may find that the strongest possible experimental result arrives too late to help the very application it was meant to support.

The “research tool” workaround and its limits

A common and often clever response to a method-of-treatment objection is to reframe the claim away from therapy and toward a non-therapeutic use—for example, the compound as a binding ligand or reference standard for screening other candidates. This can succeed in overcoming subject-matter objections, since research tool claims (molecular probes, reference compounds, and binding ligands) are recognized as patent-eligible.

But reframing the claim does not escape the evidence problem—it relocates it. If the claim now asserts that the compound is a binding ligand, then binding is the operative technical feature, and binding is exactly what the disclosure must establish. Examiners will look for actual binding data—surface plasmon resonance, ELISA, microfluidic diffusional sizing, or a screening assay using the compound as a reference or competitor ligand—not a predicted affinity. Winning the subject-matter fight while losing the sufficiency fight is not much of a win.

What this means in practice

None of this argues against computational work. In silico screening is an excellent way to decide what to test. The mistake is treating it as a substitute for testing when it comes to patenting. A few practical takeaways:

• Sequence the work before you file. Let computational screening narrow the field, then generate at least confirmatory experimental data on your lead compound before the filing date—not after.
• Match your evidence to your claim’s operative feature. If the claim turns on binding, you need binding data. If it turns on inhibition or a therapeutic effect, you need assay or in vivo/in vitro data going to that effect. Predicted values will not stand in for measured ones.
• Do not rely on filing early and supplementing later. The as-of-filing rule means a “file now, prove later” instinct is actively dangerous for activity-based claims.
• Budget for the bench. For research institutions and startups, the temptation is to protect a computational result quickly and cheaply. But a granted patent that survives examination is worth far more than an early filing that cannot.

The bottom line

Computational drug discovery earns its place at the front of the pipeline. It does not belong at the front of a patent application as the sole proof of what a molecule does. In the pharmaceutical and biotech space, examiners want to see that the claimed effect was actually observed, and they want to see it in the application as filed. Plan the experimental work into your timeline early—because in this area, the data you didn’t generate before filing is data you may never be able to use.

This post is general information on Philippine patent practice and is not legal advice. Prosecution outcomes turn on the specific facts, claims, and evidence of each application.