NeoGenelogic Patents AI for Cancer Targets in the US

⏱️ Lectura: 10 min

NeoGenelogic, a South Korean biotechnology firm, secured a US patent for an artificial intelligence technology that identifies therapeutic targets against cancer, according to a report by The Chosun Ilbo.

📑 En este artículo

1. TL;DR
2. What happened with NeoGenelogic’s patent
3. The pharmaceutical industry’s race for AI-driven therapeutic targets
4. How AI-driven therapeutic targets work
5. Impact for the industry and real limitations
6. How to verify the patent and try the tools
7. What’s next for NeoGenelogic and its competitors
8. Frequently Asked Questions
   1. What is a therapeutic target?
   2. What exactly does NeoGenelogic’s patent protect?
   3. Does an AI patent mean a treatment already exists?
   4. How can I verify the patent myself?
   5. What open tools exist to experiment with this type of model?
   6. Why does the USPTO allow patenting an AI-based method?
9. References

The announcement comes amid a global race to use AI models to accelerate the slowest phase of drug development: deciding which molecule in the tumor to target before investing years and millions of dollars in the lab.

TL;DR

• NeoGenelogic secured a US patent for its AI that identifies therapeutic targets against cancer (The Chosun Ilbo).
• The South Korean company joins a growing group of biotech firms using AI to shorten the selection of therapeutic targets.
• Patenting AI in biotechnology requires a concrete practical application, not an abstract formula, according to USPTO criteria.
• Open tools like RDKit and DeepChem make it possible to reproduce parts of the pipeline that prioritizes drug candidates.
• AlphaFold, from Google DeepMind, proved that AI predicts protein structures with precision useful for research.
• The patent doesn’t replace experimental validation; the AI candidate still needs to be tested in the lab before a clinical trial.
• Anyone can verify the patent by searching for NeoGenelogic as the assignee on Google Patents or USPTO Patent Public Search.

What happened with NeoGenelogic’s patent

The United States Patent and Trademark Office (USPTO) granted NeoGenelogic exclusive rights to an artificial intelligence method designed to identify therapeutic targets in cancer, according to the report by The Chosun Ilbo. In biotechnology, a therapeutic target is the molecule, almost always a protein, that a drug is designed against to block or modify its function in the tumor cell.

The original report doesn’t detail the patent number or the exact molecular mechanism the registration protects, which is common in general press coverage of technical patents. Further below, in the section on how to verify the patent, we explain how any reader can search the full file using the company name as the assignee.

What matters isn’t just the grant itself, but what it represents. A South Korean biotech firm managed to protect, in the US market (the largest in the world in pharmaceutical spending), an AI method applied to oncology. That gives NeoGenelogic a legal edge over competitors who want to commercialize a technically similar approach in the US during the patent’s term.

The pharmaceutical industry’s race for AI-driven therapeutic targets

Identifying a valid therapeutic target is, historically, the most expensive bottleneck in drug discovery. Before synthesizing a single candidate molecule, labs need to confirm that blocking a specific protein actually slows tumor growth without damaging healthy tissue. That process, backed by experimental screening and genomic studies, can take years and consume a large share of a research program’s budget.

The rise of deep learning models changed the starting point. AlphaFold, developed by Google DeepMind, showed that a neural network can predict a protein’s three-dimensional structure from its sequence with enough precision to guide real research, not just publish an academic paper. That result opened the door for companies like Isomorphic Labs, DeepMind’s drug discovery spinoff, and a wave of Asian and American startups to build full AI pipelines to prioritize candidates before touching a test tube.

NeoGenelogic joins that trend with its own oncology-focused approach, at a time when several biotech firms are looking to differentiate themselves through intellectual property over their AI models, not just over the final molecules that eventually reach a clinical trial.

How AI-driven therapeutic targets work

Although NeoGenelogic hasn’t published details of its architecture, systems of this kind share a general structure well documented in the bioinformatics literature. The typical pipeline combines gene expression data from tumor and healthy tissue, protein-protein interaction networks, and predictive models that score each candidate protein by its “druggability”: how feasible it is to design a drug that blocks it selectively.

The following diagram summarizes that workflow at a high level, from raw data to the patent application:

flowchart TD
    A["Omics data: gene and protein expression"] --> B["AI model: prioritizes candidate proteins"]
    B --> C["Structure and affinity prediction"]
    C --> D["Experimental validation in the lab"]
    D --> E["Patent application for the method"]

In practice, a researcher wanting to reproduce the first stage of that pipeline, calculating a candidate’s molecular properties, can do so today with open libraries. A minimal example with RDKit, the most widely used open source cheminformatics library in the field:

from rdkit import Chem
from rdkit.Chem import Descriptors

molecula = Chem.MolFromSmiles("CC(=O)Oc1ccccc1C(=O)O")  # aspirin, as an example
peso_molecular = Descriptors.MolWt(molecula)
logp = Descriptors.MolLogP(molecula)

print(f"Molecular weight: {peso_molecular:.2f}")
print(f"LogP (lipophilicity): {logp:.2f}")

That snippet calculates two properties that any “druggability” model uses as input: molecular weight and LogP, a measure of lipophilicity that influences whether the compound can cross cell membranes. A real target identification pipeline chains together dozens of properties like these along with binding affinity predictions, differential expression in the tumor, and estimated toxicity, to produce a ranking of candidates.

def puntuar_diana(expresion_diferencial, afinidad_predicha, toxicidad_estimada):
    # illustrative weights; each lab adjusts its own
    score = (0.4 * expresion_diferencial
             + 0.4 * afinidad_predicha
             - 0.2 * toxicidad_estimada)
    return score

candidatos = {
    "PROT_A": puntuar_diana(0.82, 0.75, 0.10),
    "PROT_B": puntuar_diana(0.55, 0.90, 0.35),
}
mejor = max(candidatos, key=candidatos.get)
print(f"Prioritized target: {mejor}")

This second block is deliberately simplified: it illustrates the logic of combining signals (differential expression, affinity, toxicity) into a single score, the type of method usually described in a patent application like NeoGenelogic’s, although the actual weights and variables of each company are protected intellectual property.

💭 Key: what’s usually patented isn’t “using AI for cancer” in the abstract, something the USPTO would reject as an abstract idea, but the concrete method: what data is combined, how the model is trained, and what specific practical application it solves.

Impact for the industry and real limitations

The following table compares NeoGenelogic’s approach, without knowing its exact implementation, with the most established methods for finding a therapeutic target against cancer:

Approach	When it’s used	Advantage	Limitation
In vitro experimental screening	Early validation of an already suggested candidate	Direct biological evidence, high reliability	Slow and expensive: compounds tested one at a time
Functional genomics (CRISPR screens)	Discovering essential genes in a tumor type	Broad coverage of the full genome	Requires specialized lab infrastructure
Classic molecular docking	Estimating protein-drug binding affinity	Based on physical principles, interpretable	Expensive at scale and with limited precision
Predictive AI models (like the one patented by NeoGenelogic)	Prioritizing candidates before entering the lab	Drastically reduces the initial search space	Depends on the quality and bias of the training data

The point that often gets lost in coverage of this kind of announcement is that a patent protects a method, not a clinical outcome. NeoGenelogic secured exclusive rights to its way of identifying targets, but that doesn’t mean a drug is already in development, nor that the molecule eventually chosen will make it through clinical trial phases, a process that in oncology usually takes several additional years and where most candidates fail.

⚠️ Watch out: an AI patent doesn’t certify clinical efficacy. It legally protects the target selection method; validation in humans remains a separate process, regulated by agencies like the FDA, and statistically most oncology candidates never get approved.

How to verify the patent and try the tools

Any reader can confirm the patent’s status without relying on press coverage. The Google Patents search lets you filter by assignee by typing assignee:NeoGenelogic into the search bar, and the official USPTO Patent Public Search system accepts the field code AN/NeoGenelogic to restrict the search to the company name as assignee.

To explore the kind of open source tools that power these pipelines, installation takes just a couple of minutes:

# basic cheminformatics
pip install rdkit

# deep learning models for drug discovery
pip install deepchem

With those two libraries installed, the RDKit snippet from the previous section already runs locally. DeepChem adds pretrained models to predict solubility, toxicity, and binding affinity, the same signals a target identification model needs to combine.

💡 Tip: the EBI AlphaFold Protein Structure Database freely publishes millions of predicted protein structures, useful as a starting point to explore candidates without relying on your own computing infrastructure.

What’s next for NeoGenelogic and its competitors

The logical next step, and the one that will determine whether the patent has real commercial value, is whether NeoGenelogic or a pharmaceutical partner advances a concrete candidate toward preclinical studies. That’s usually when information that’s currently missing becomes public: the target cancer type, the specific protein, and the first efficacy data in animal models.

Meanwhile, other Asian and American biotech firms will likely keep filing similar applications with the USPTO, in a race where intellectual property over the AI method is becoming as contested as intellectual property over the final molecule. Whether the office tightens or loosens its eligibility criteria for this kind of hybrid invention, between software and biotechnology, in upcoming filings remains to be seen.

📖 Summary on Telegram: View summary

Try it yourself: install RDKit with pip install rdkit and calculate the molecular properties of a real compound in under ten lines of Python.

Frequently Asked Questions

What is a therapeutic target?

It’s the molecule, almost always a protein, that a drug is designed against to block or modify its function inside a cell, in this case a tumor cell.

What exactly does NeoGenelogic’s patent protect?

According to the report by The Chosun Ilbo, it protects the artificial intelligence method the company uses to identify these targets in cancer, not a specific molecule or drug already developed.

Does an AI patent mean a treatment already exists?

No. The patent covers the selection method; the identified candidate still has to go through experimental validation and, eventually, regulated clinical trials before becoming an approved treatment.

How can I verify the patent myself?

By searching for “NeoGenelogic” as the assignee on Google Patents or using the field AN/NeoGenelogic on USPTO Patent Public Search.

What open tools exist to experiment with this type of model?

RDKit for basic cheminformatics, DeepChem for deep learning models applied to drugs, and EBI’s public AlphaFold structure database to start from already computed protein predictions.

Why does the USPTO allow patenting an AI-based method?

Because, unlike an abstract idea, the method describes a concrete practical application: what data it combines, how it processes it, and what technical problem it solves, the criteria the office requires under its eligibility framework for software and biotechnology inventions.

References

• The Chosun Ilbo via MSN: original report on NeoGenelogic’s patent for its AI technology against cancer.
• USPTO: United States patent and trademark office, the official source to verify the status of any application or granted patent.
• Google Patents: public search engine that lets you filter patents by assignee company, including NeoGenelogic.
• RDKit: official documentation for the open source library used in this article’s cheminformatics examples.
• Wikipedia: AlphaFold: background on Google DeepMind’s protein structure prediction system.

📱 Do you like this content? Follow @programacion on Telegram for daily tech content in Spanish: quick summaries, fresh content every day.

Imagen destacada: Foto de National Cancer Institute en Unsplash