Every drug-whether a tiny chemical compound or a complex antibody-works by latching onto its specific target molecule and changing its behavior in the body. Identifying the target, therefore, is the crucial first step in creating a new drug. But how do we actually determine which molecule to target?

History of Target Molecule Hunting

For the last fifty years, drug discovery has largely focused on uncovering the molecular targets responsible for disease. Methodologically, this pursuit can be divided into three major eras, each shaped by advances in technology and scientific insight.

1975-2000: The Era of Classical Molecular Biology

Advances in gene cloning and genetically-engineered mouse models transformed this era, enabling researchers to probe what individual genes are doing in our body. In diseases characterized by accumulation of abnormal proteins, defining those proteins became a central research priority. With these discoveries, researchers hypothesized which molecules might be driving disease, and therefore could serve as therapeutic targets. Major targets such as PD-1, HER2, and amyloid-beta were defined during this era using classical molecular biology approaches. It was also a time when ‘serendipity’ was a buzzword, reflecting how unexpected finding often guided discovery.

2000-2015: The Era of Genomics

During this period, advances in the human genome decoding drove rapid progress in identifying disease-related molecular targets. In cancer research in particular, pinpointing cancer-specific genetic abnormalities became the leading strategy for drug-target discovery, yielding key targets such as KRAS, EGFR, BRAF and ALK.

Today, however, the discovery of new driver mutations has slowed, suggesting that genomics-based target identification is reaching a plateau. As a result, fewer startups now focus solely on genetic abnormalities for drug-target discovery.

2010-2025: The Era of Proteomics and Transcriptomics

These new technologies have taken on a central role in disease characterization, enabling researchers to detect abnormal peptides, to map protein- and gene expression patterns at an individual cell level. Transcriptomics especially advanced rapidly with the rise of single-cell analysis. This has made it possible to measure subtle expression differences cell by cell, revealing far more detailed views of tumor heterogenicity, immune-cell interactions and cellular transition during disease progression. Together, these research fields offer substantial potential for future drug-target discovery, expanding beyond what genomics alone can provide.

The New Trend of AI-driven Drug Discovery Since 2020

Around 2020, AI-driven drug discovery emerged as a new direction in biomedical research. Breakthrough such as AlphaFold from Google DeepMind have accelerated interest in applying AI to biological problems. It’s important to note, however, that while AlphaFold does excel at predicting protein structures, it is not designed to identify drug targets. The main limitation is the lack of large, high-quality datasets describing disease-specific protein alterations, which are essential for training AI system to recognize potential targets. In short, AI-based protein structure prediction offers enormous promise, we still remain in early stages in solving the fundamental challenge of how to figure out drug-targets.

Above is the brief review of the target-molecule hunting history.

Target-molecule Discovery: The Most Critical yet Challenging Task

Identification of target molecules remains the most critical yet the most challenging task in new drug discovery. Over the last fifty years, researchers worldwide both in academia and pharmaceutical industry have devoted enormous time and resources into this pursuit. Yet despite these efforts, only a small number of candidate molecules have ultimately been translated into real medicines.

Infectious diseases allow easier identification of drug-targets because they are caused by specific external pathogens. In contrast, cancer, degenerative diseases, and autoimmune disorders pose far greater challenges, as the mechanisms underlying their onset and progression are considerably more complex. What makes this matter more challenging is that, even with the same disease, the target can vary from patient to patient.

Failing to deliver proper medical care for patients with difficult diseases-conditions referred to as Unmet Medical Needs-ultimately stems from this dauting task of drug-target discovery. No matter how demanding the task might be, however, we must continue pushing forward in our search for new drug-targets, driven by the immense business, scientific and humanitarian potential in this field. Alongside COGNANO, two other ventures have taken on this ambitious mission of AI-driven drug-target discovery. Today, I’m honored to introduce these pioneering startups – Recursion and Insilico Medicine-and share what makes their work so exciting.

Trailblazers in AI-driven Drug Discovery

Recursion, based in Salt Lake City, Utah, is an AI-driven drug discovery company. It’s distinctive platform, Recursion OS, combines cell imaging, perturbation experiments, and deep learning to map disease biology, identify targets, discover candidate compounds and advance clinical development. The company’s core strength lies in its AI-powered analysis of relationships among diseases, genes and compounds, enabled by an extensive cell-phenotype dataset.

Insilico Medicine is a generative AI-driven drug discovery company built around Parma AI, an integrated platform that spans targets identification, compound design, and both preclinical and clinical development. The system unifies disease-specific target discovery, low-molecular-weight compounds design into a single workflow. The company also conducts clinical trials for its own AI-designed candidates.

Business Models Common to The Two Ventures

Although these ventures take different technical approaches to AI-driven drug discovery, they share some important characteristics.

First, their founders and core research teams possess deep expertise not only in conventional biology but also in mathematics, information science, AI, and data science. They view targets and molecule discovery as a problem rooted in mathematical and informational structures, aiming to redesign the early fundamental phase of drug discovery through AI.

Second, they generate large-scale proprietary datasets rather than relying on external sources. Their AI analyze these internally generated datasets to explore potential targets and compounds. Instead of focusing on AI software development, these companies emphasize the capacity to produce high-quality, high-volume biological data in building competitive edge in drug discovery.

Third, these companies adopt an integrated business model that spans the entire drug development chain-from early discovery seeds to lead generation and ultimately preclinical and clinical development.

Fourth, they manage every step required to bring a new drug in market, from initial target identification to clinical development and regulatory approval. This end-to-end approach requires heavy investment in research and development, clinical trials, and overall operations, making the business highly capital-intensive and dependent on Nasdaq―listing and large-scale financing.

Why are these ventures so exclusive to their own pipelines?

In theory, AI and large-scale data could generate an endless stream of new drug targets. In reality, however, limited capital and human resources restrict how far the ventures can expand their pipelines on their own. Why, then, do these companies remain committed to advancing their own pipelines rather than simply licensing out AI-generated candidates? Several factors appear to drive this highly-exclusive business approach.

First, a major reason is the high failure rate of small-molecule candidates through clinical development. Providing AI-generated candidates alone is not enough to attract external partners. Potential business partners need more than computational prediction-evidence that a candidate has a real therapeutic promise. Only by advancing a candidate through animal studies, toxicity testing, and early clinical trials, the ventures can offer credible proof of efficacy and safety, making the assets far more compelling to collaboration or licensing.

Second, the companies treat information on target molecules as a core confidential asset. Any leak regarding the molecules or their intended disease targets could prompt their competitors to pursue the same targets-especially now that AlphaFold is widely available. As a result, these ventures tend to keep their research highly confidential until they have validated the molecules, secured patents, and advanced the candidates through preclinical development.

Third, in small-molecular drug discovery, there’s an inherent linkage among the target molecule, its mechanism of action, the compound’s structure, and the associated use patent. The tight interdependency makes it difficult to divide development rights among multiple business partners. As a result, the ventures tend to pursue an integrated pipeline to capture as much value as possible from their assets.

Fourth, ventures want to keep information about newly-identified targets confidential for as long as possible. As a result, they tend to maintain a low profile throughout patent applications, publications, and early clinical development. This secrecy may make sense from an asset-protection standpoint, but it also discourages open collaboration with external researchers, leaving the company relatively isolated.

What does COGNANO have in common with Recursion and Insilico Medicine?

COGNANO shares the goal of discovering new target-molecules with these ventures. Like these companies, we use large, internally-generated biological datasets and machine-learning methods to extract disease-relevant insights. Unlike them, however, COGNANO prioritizes open collaboration: we engage external researchers, freely provide our antibodies, and value third-party validation of our data. This difference in approach reflects the unique characteristics of the targets we work with.

Why does COGNANO value – and in fact need – collaboration with external partners?

First, COGNANO’s goal is to detect disease-specific structural change in a molecule, rather than the molecule’s baseline form. A large protein in cancer cells, for example, can adopt a shape that differs markedly from its counterpart in normal cells due to truncation, glycosylation, complex formation, localization shifts, and conformational alterations. Our VHH antibody functions as a probe that recognizes these pathology-specific structural states, not just the canonical amino-acid sequences.

Second, current algorithm such as AlphaFold cannot predict pathology-specific structural changes. While AlphaFold accurately models a protein’s baseline form from its amino acid sequences, it cannot capture context-dependent alterations, including cancer-specific posttranslational modifications, truncations, membrane-associated conformations, dynamic complex formations or microenvironment-driven structural state. Thus, identifying target molecules alone is insufficient for competitors to reproduce our pipeline.

Third, our core intellectual property goes beyond VHH antibodies themselves. It includes our unique ability to detect disease-specific antigen conformations and the corresponding VHH datasets that recognize them. In contrast to companies like Recursion or Insilico Medicine, whose principal assets are newly-discovered antigen or target annotations, our competitive advantage lies in the precise identification, quantification, and reliable validation of disease-relevant three-dimensional antigen structures.

Fourth, elucidating disease-relevant antigen structures requires a uniquely integrated platform: mass VHH libraries generation, pathological cell and tissue screening, NGS, and machine learning-driven analysis. Even if the target is publicly available, no one can match COGNANO’s capabilities without building a robust pipeline capable of producing antibody datasets that truly capture disease-specific structures.

Fifth, COGNANO sees a strong need to expand global collaboration to further validate and advance our VHH antibodies and our datasets on diverse disease-specific structures. Rather than keeping every achievement to ourselves, we believe it is far more valuable to establish the clinical significance of our targets by offering antibodies and datasets for research purposes, and actively collaborating with academia, diagnostic, and pharmaceutical industry worldwide.

The Essence of COGNANO’s Model

To wrap up, companies such as Recursion and Insilico Medicine can be described as AI-native chemical drug discovery companies. They use AI to identify drug-targets, and then design and develop small-molecule compounds against those targets internally.

In contrast, COGNANO positions itself as a structure-native target discovery company. Our platform detects aberrant molecular structure in pathological tissue using VHH antibody big data and machine learning.

I’d like to stress the point once again that COGNANO is not searching for the target molecule itself. In traditional drug discovery, molecules such as HER2, PD-1, or KRAS were treated as drug-targets in their baseline forms. In real pathological tissue, however, these molecules can adopt disease-specific conformational states that differ from their counterparts in normal tissue. COGNANO’s focus is on detecting these aberrant, disease-associated structural states. In other words, we aim to identify the structural transition that a molecule undergoes during disease progression. Those differences in the approach to drug discovery naturally led to different business models.

For companies like Recursion and Insilico Medicine, target discovery and compound design are deeply interconnected processes. This makes an internal pipeline-driven drug discovery model a logical business strategy. As a result, these companies focus on building the value of their own pipelines while simultaneously pursuing collaboration with pharmaceutical partners.

In contrast, COGNANO continues to produce assets of disease-specific molecular structures. These assets can be applicable to a broader range of use cases, including diagnostic markers, patient-stratification markers, ADC targets, antibody-based drug discovery, as well as drug-target identification.

COGNANO has already identified multiple novel structural targets that are conserved across diverse human cancers, paving the way for development of broad-spectrum anti-cancer therapeutics. To facilitate collaboration, we have initiated a program offering eight pancreatic-cancer-specific antibodies, which are freely available for research use. Anybody interested in the antibodies is welcome to visit the URL below for access. Details regarding their specificities are presented in AACR2026 for further review.

The concept of daughter ventures plays an important role in our future business strategy. It is not realistic for a single company to manage multiple disease-specific structures, because the required expertise varies significantly from one disease area to another. For example, brain cancer relies on a completely different network of researchers, clinicians and investors than pancreatic cancer. To address this, we believe it is more effective to divide our pipeline into two specialized groups:

1. Target discovery-responsible for identifying structure-targets
2. Therapeutic development-responsible for advancing those targets into real medicines

Under this model, COGNANO will function as a Target Operating System (Target OS)-a foundational engine for drug-target discovery. Each validated target can then serve as a basis for launching an independent daughter venture.

Daughter ventures bring together universities, hospitals, investors, CROs and pharmaceutical companies to form disease-specific teams optimized for managing each condition. COGNANO contributes disease-specific structures to each daughter venture, and receives equity stakes and royalties in return. This business model differs fundamentally from the conventional venture approach, where investment is funneled into a single, internally-controlled drug discovery pipeline.

Our strength does not come from keeping newly-discovered targets secret or monopolized. Instead, it lies in our unique ability to detect disease-specific structural states-a capability that goes beyond what traditional genomic decoding and conformational prediction technologies can achieve. Equally important is our ability to convert those structural-findings into validated asset through VHH antibody probes, AI-driven analysis, and clinical studies. This approach brings us far closer to the true goal of drug discovery than simply identifying novel targets. In practice, only the well-orchestrated integration of VHH antibody probes, screening systems, NGS data, AI interpretation, and corresponding clinical data creates a real, defensible assets. In that sense, COGNANO’s competitive advantage stems not from any single antibody or target, but from our platform itself-its ability to discover, interpret, and substantiate disease-specific structural information.

We call this model Target OS. Our revenue base spans VHH antibody distribution for research, data licensing, biomarker businesses, drug-target licensing, equity stakes from daughter ventures, and downstream royalties. But our ultimate ambition is larger than becoming another pharmaceutical powerhouse. We want to be an engine of wisdom-a source of structural insight that sparks new ideas, new medicines and new companies.

By sharing disease-specific structures with researchers, clinicians, and entrepreneurs worldwide, COGNANO aims to ignite breakthroughs in diagnostics, therapeutics, and to accelerate the emergence of new pharmaceutical leaders.

Target OS is our way of shaping what the drug business becomes next.

This is edited by Dr. Kimio Fujii for english.