103 · L5 IP Litigation & Disputes in Biotech

Overview — the master logic of biotech IP disputes

This lecture answers a different question from the previous four. L1–L4 asked how do you obtain and use IP. L5 asks what happens when someone contests it. The through-line is that a granted patent is not a right, it is a claim — and every granted claim is assumed false until it survives challenge. In biopharma that assumption bites harder than in most industries, because the same product can simultaneously be attacked on eligibility, on written description, on infringement and on regulatory ground. The lecture runs in three parts: Part 1, the forums — who hears what, and how the US and Chinese systems are structurally different; Part 2, the six recurring types of biotech dispute with their case law; Part 3, the emerging trends — AI as inventor, the complexity of modern biologics, and platform patent battles such as CRISPR.

Key Terms 關鍵術語

Part 1 — Litigation framework: the two forums

The US court system 兩制法院架構

Understanding the US system matters because the deck's case law is US case law. The structure is a two-branch design: an administrative branch and a judicial branch.

One point that is easy to miss and easy to be examined on: the Federal Circuit is not a typical appeals court. It is a court of appeals with nationwide jurisdiction over patent appeals, and its decisions bind every district court. That is why a Federal Circuit ruling such as Thaler v. Vidal on AI inventorship effectively sets the national rule overnight.

The Chinese IP court system 中國知識產權法院體系

China rebuilt its patent litigation architecture from scratch in three waves, and the dates matter because they explain the structure:

• 2014 — Beijing, Shanghai and Guangzhou set up IP Courts as first-instance venues for patent cases, sitting alongside the provincial High Courts.
• 2017–2018 — eighteen local IP Tribunals created in Chengdu, Wuhan, Hefei, Fuzhou, Xi'an, Nanchang, Nanjing, Suzhou, Zhengzhou, Tianjin, Changsha, Changchun, Hangzhou, Ningbo, Jinan, Qingdao, Shenzhen and Lanzhou, plus the Beijing Internet Court.
• 2018 — the Supreme People's Court established IP Tribunals to handle all second-instance and final patent cases, creating a unified national appellate body.
• CNIPA sits outside the court hierarchy entirely — it examines, reexamines and invalidates patents administratively.

Litigation vs. Invalidation — the core distinction

Note the strategic consequence: because the defendant can raise invalidity as a counter-suit, a patentee suing for infringement is automatically exposing the validity of the very patent it relies on. Most infringement settlements are therefore priced off the combined risk of both questions, not just one.

Unified vs. bifurcated — why the forum split matters strategically

Part 2 — The six types of biotech dispute

The deck organises the whole field into six recurring fights. Each is worth being able to name, define and give a case for.

Type 1 — Patent eligibility of nature

Natural products, natural phenomena and abstract ideas sit outside patentable subject matter. The practical question for a biopharma company is always the same: does the claimed thing sit on top of nature, or has it been made to differ from nature enough to escape the exclusion?

Type 2 — Claim breadth, written description and enablement

The scope triangle

Claim scope is a three-way trade-off, and drafting is choosing where to sit on it:

• The claims did not describe the antibody's structure — only what it binds to. Functional claims are not enough.
• Only 26 examples were disclosed, when millions and millions of possible sequences existed — a finding of undue experimentation.
• The standard moved permanently. The takeaway slide is blunt: the standard for enablement and written description is much higher now, and the old “newly characterised antigen” route is closed.

Type 3 — Judicial exceptions: abstract ideas

The second statutory exclusion, alongside nature, is the abstract idea. The enumerated categories are:

Type 4 — Patent linkage / Hatch-Waxman litigation

How the mechanism works

The system is best understood as a pre-launch gauntlet. The regulator publishes an “Orange Book” of patents covering the approved reference drug; the generic applicant must certify its position against every one of them before it can get marketing approval; if the generic says a listed patent is invalid or not infringed, the patent owner has a short window to sue; and the litigation outcome — not the regulator — decides whether the generic enters the market.

The four generic certifications

What happens after the litigation

The China slide adds the incentive that makes the system work: the first generic to successfully challenge gets 6 months' data exclusivity and 12 months' market exclusivity. That is the reward for being first, and it is why the category analysis is a race rather than a checklist.

China's first patent linkage decisions

• Roche v. Chugai (CN1938034B) — a composition of eldecalcitol with an antioxidant, where the antioxidant is added to prevent degradation to specified metabolites and the level stays at 1% or less after 12 months. The IP Court ruled against Roche.
• Purdue (three asserted patents) — CN105267170B and CN102657630B both defined oxycodone HCl formulations by polyethylene-oxide molecular weight and rheology; CN101812065B claimed an oxycodone extended-release matrix of ≤25 ppm 14-hydroxycodeine ketone. The IP Court ruled against Purdue.

The technical lesson from the Purdue set is worth extracting: patenting on a measurement window (molecular weight, rheology, ppm threshold) produces claims that are numerically precise yet jurisdictionally fragile, because the measuring method is exactly what the accused party can contest.

Type 5 — Biosimilars and the patent dance

Biologics follow a different route from small molecules because they are approved through a BLA rather than an NDA, and because the reference product sponsor is called the RPS rather than an NDA holder.

• 180-day notice. If a biosimilar is to enter the market, the manufacturer must notify the RPS no later than 180 days before the commercial marketing date, including its process patents (composition of matter, formulations, medical use, extraction methods for TCMs) where an NDA/BLA is on file.
• The dance is optional. Established in Sandoz v. Amgen, the patent dance is a voluntary process designed to streamline litigation.

1. Biosimilar applicant initiates — provides its abbreviated BLA (aBLA) and details of its manufacturing process to the RPS.
2. RPS responds — provides a list of patents it believes could be infringed, plus an initial licensing offer.
3. Biosimilar applicant responds — explains why its product does not infringe, or why the patents are invalid or unenforceable.
4. Negotiation — a series of information exchanges over several months, producing a final list of patents that may be litigated immediately (Phase i) versus those that could be litigated later, closer to launch (Phase II).

Type 6 — Skinny label litigation

A skinny label is a regulatory carve-out: the generic or biosimilar application removes the patented indication while keeping the unpatented ones, so the label no longer reads on the protected use.

The purpose is market entry for the unpatented uses. The relevance is to biosimilars with multiple indications. The risk is induced infringement if the marketing implies the patented use.

Part 3 — Evolving trends

Trend 1 — AI and drug discovery

Four questions arise: can AI be an inventor, who owns an AI-made invention, and will patent offices accept in silico data in place of real experimental data.

How do you patent an AI invention?

The Chinese framing on the slide is the most useful one for this course. An AI algorithm is not patent-eligible on its own. It becomes so when you add a technical solution:

The worked example — Google's Transformer

The slide reproduces the opening of the claim for a sequence-transduction neural network using self-attention (encoder with self-attention sub-layers, decoder, encode-and-generate), and then states the outcome: ALLOWED in the US, REJECTED in China on appeal (rejection decision 201880007309X). The same claim, two offices, opposite results — the cleanest illustration that AI patentability is jurisdiction-specific.

• Improving the computer itself. Show how the algorithm directly improves the computer's functioning; ideally at the hardware level, showing hardware resources being specifically scheduled, reduced or improved.
• Improvements for specific applications. Draft the practical application into the claim as filed (draft many different claims); provide examples and detailed description of how the algorithm interacts with technical features or specific data to produce specific technical effects; create clear steps tied to actual tangible inputs (natural language text, data); the more specific, and the more different examples, the better; explain the technical effect in the specification.
• User experience. Where relevant, show how the algorithm interacts with technical features to improve the user experience.
• Inventive step. Show how the algorithm causes the improved technical effect.

Other impacts of AI on patents

• Standard for inventiveness. Who is the “skilled person in the art” now? Is the inventiveness standard higher, since we are all “smarter” with AI? The obvious extension of the obviousness question.
• Is AI legitimate prior art? Some are already trying to “pollute” the prior art to prevent others protecting new inventions.
• Data protection. The massive datasets used for AI training are very valuable, and how best to protect them is an open question — patents, trade secrets, or copyright? Artists are already suing AI companies; if massive patient data is used, will patients do the same?

Trend 2 — biologics are more complex, so IP protection must be too

Before, the patentable unit was a small molecule or a biological sequence. Now the unit is a combination of components, and the claim has to describe each of them:

The generative list of what must now be claimed in parallel: ADCs, gene therapy and gene-editing tools (CRISPR Cas9), cell therapies (CAR-T), mRNA and siRNA, and drug-delivery platforms (LNP).

Trend 3 — global protection strategy

• Coordinate patent protection strategy around the world rather than jurisdiction by jurisdiction.
• Regulatory rules differ, and they favour whoever enters earlier — China gives PTE and possibly full data exclusivity to those who first file in China.
• Genetic information privacy laws differ in each jurisdiction.

Trend 4 — platform patent battles: the CRISPR story