Using AI to select the right disease model for cancer drug development


Cancer researchers are constantly looking for disease models that closely resemble human physiology to test for cancer drugs. Artificial intelligence (AI) can help researchers sift through the vast amounts of available data, saving time and money, to find the most suitable disease model for oncology drug development.

In the early stages of drug development, the choice of disease model used for drug discovery and drug testing can make or break a study. The drugs are only approved for studies in humans based on satisfactory safety and efficacy results from experiments conducted using specific models. This ultimately leads to market authorizations for the drug’s marketing and helps deliver an optimized product for patients in need.

Using the “wrong” model can be costly for a drug discovery project for a number of reasons. First, the disease models themselves can be expensive to find, leaving scientists with less margin for error in selection, as study budgets are often limited at this point.

Second, the characteristics of the chosen model, including specific mutations or biological pathways, can strongly influence the effectiveness of a drug. Even promising drug candidates can provide conflicting results between different models, causing some drugs to fail in human clinical trials despite passing preclinical tests.

Therefore, using a relevant disease model can dramatically improve the credibility of research as well as its results.

A world of options: from 2D in vivo models to 3D in vitro models

Today, there are many types of preclinical models that researchers can exploit, each with their own experimental, monetary, and logistical advantages.

Javier Pineda, Data Scientist at

Traditionally, immortalized cell lines have been the most common tool used to test potential drug targets in oncology research. These are human cells that have been manipulated to multiply indefinitely in the laboratory.

Based on the same principle of testing drugs on cultured cells, scientists also use primary cell lines derived from patients. Taken from patients, these cells provide a better 2D representation of the patient’s actual physiology.

There are also 3D models, such as spheroids or organoids, which are closer to the physiology of the patient than individual cells. These models consist of a simplified version of an organ or tumor produced in vitro and aim to simulate the immediate cellular environment of tumor cells in a patient.

An improvement over cultured cell models are in vivo animal models, which capture the environment of tumor cells at the organism level. A vivid example are mouse models, which have been integral to the discovery of cancer drugs for decades. In fact, syngeneic mice are immunologically compatible and have been selected for drug testing.

“Syngeneic mouse models and patient-derived xenografts (PDX) are popular in vivo models in oncology ”, said Javier Pineda, Data Scientist at, an e-commerce platform that connects buyers and sellers of personalized research services.

“PDX models involve implanting cancerous tissue from the patient into an immunocompromised mouse, so that the tissue is not rejected by the mouse’s immune system.”

PDX models are representative of the evolution towards personalized medicine. They allow the most appropriate therapeutic intervention to be selected by truly capturing the 3D environment (albeit without the functioning of the immune system) in the patient.

More recently, as an improvement of PDX, humanized mouse models with reconstituted human immune systems have also been developed.

Obstacles to choosing the right disease model

With a large number of preclinical disease models available, it can be quite difficult to choose the most suitable for a cancer study.

Even knowing how to get a model can be a challenge to get started, Pineda said. For a single study, researchers routinely end up contacting multiple vendors to inquire about model availability. This often results in a trial and error process that inevitably takes time.

Once a supplier has been identified, accessing relevant data that will aid in decision making is another hurdle. Vendor websites and catalogs typically only provide annotation-level data.

This refers to critical or explanatory notes provided on disease models and includes information such as tumor type and subtype, mouse strain type, geographic location of the disease model, and background information. on the patient.

“It is imperative to study factors such as genetic mutations, target gene expression or drug sensitivity before looking for a disease model. Data at the annotation level alone is generally insufficient. Even in cases where molecular data is available, due to its proprietary nature, it is often inconsistent between providers and difficult to compare ”, Pineda explained.

These limited data prevent researchers from performing the necessary quality checks as well as making quantitative comparisons between models provided by different vendors. Consequently, the choice of the model is strongly restricted, which can influence the result of the experiments carried out.

Finally, negotiating a project proposal, finalizing a contract, and managing regulatory approvals can also be difficult. For some companies, this process takes weeks or even months, resulting in delays in project schedules.

Disease Simplified model selection: AI-driven decision making

Based on user feedback on their platform, the team realized the need for a centralized, data-driven approach to researching disease patterns. This triggered the project of Disease pattern research, an AI-based tool that allows researchers to perform quantitative model comparisons between vendors by leveraging machine learning algorithms.

“Our existing relationship with many disease model vendors has enabled us to consolidate molecular data from all vendors while ensuring data confidentiality and security.” Pineda said.

“The researcher uses a variety of algorithms to process, aggregate and visualize data from large datasets based on molecular sequencing. To efficiently find and search for suitable disease models, the tool allows researchers to filter by mutations, gene expression, cancer types, etc.

Often, studies require a panel of disease models to test the implications of a target drug on multiple types and subsets of cancer. To facilitate the selection of multiple models, the Disease Model Finder has a compare function that provides information on models that may be biologically distinct or similar depending on the requirements of the study.

The team behind the Disease Model Finder sought to make compute-based comparisons accessible to those without a background in bioinformatics.

“The objective here is to make the analysis user-friendly so that researchers can improve their research while saving time, without having to resort to external expertise” Pineda explained.

Disease Model Finder,, Cancer Drug Discovery, AI
The Disease Model Finder allows researchers to perform differential gene expression (DGE) analysis (top left), as well as visual comparisons of disease models versus clusters of models generated by RNA-Seq (top right). Aggregation filters allow researchers to sift through thousands of disease models (bottom).

In the event that a researcher cannot find a model that relates directly to their study, the Disease Model Finder allows them to create a custom request. This is sent to the platform’s Research Concierge team, which can then identify the suppliers most likely to have the model in question.

The Disease Model Finder also works in conjunction with COMPLi from® functionality, which covers compliance processes when researching disease patterns and other regulated services.

“This configuration ensures supplier compliance, enables automated purchasing and reduces processing time, allowing researchers to do everything using a single platform.” Pineda said.

The evolving role of AI in oncology research

The Disease Model Finder currently hosts a range of PDX models, but will be expanded to include more types of oncology models in the coming months. In addition, the team also plans to add more bioinformatics tools, such as a recommendation algorithm that uses correlation analysis to suggest disease models of interest.

With the increased use of AI in disease research, including oncology, Pineda said the world can expect interesting applications such as AI-based drug response predictions. very soon become a reality.

“Predicting how a patient will react to therapy is a key line of research in the field of oncology. This requires analyzing large sets of drug response data and other molecular information ”, He continued.

“Incorporating drug response predictions into the Disease Model Finder would dramatically improve the search for cancer models. As a constantly evolving platform, this is a feature that we hope to offer our researchers in the future ”, concludes Pineda.

To learn more about the AI-powered disease model finder and how it can aid your drug discovery research, visit company website. If you are interested in partnering with and providing scientists around the world with access to your disease models, contact us via

Images courtesy of

Source link


Comments are closed.