WEBINAR: AI and HCS in Cancer Research: Case Studies on Faster Target Discovery and Screening

AI and HCS in cancer research: Facilitating faster target discovery and screening

Last week Ardigen co-hosted a webinar together with our partner Ryvu Therapeutics on the topic of applying AI and High-Content Screening (HCS) in cancer research. Experts from Ryvu (Andrzej Mazan and Miika Ahdesmäki) and Ardigen (Krzysztof Rataj and Magdalena Otrocka) highlighted the ways HCS in combination with AI for data analysis can help facilitate breakthroughs in cancer research and shared recent case studies of successfully using these methods in target discovery and screening projects. 

The two teams shared real-world examples of how they are using HCS data in combination with AI-powered insights for target discovery and screening, including 1) identifying specific targets for colorectal cancer, 2) using deep learning to predict EdU staining at a much lower cost, and 3) implementing quality control for HCS data analysis. You can watch the full webinar on-demand or keep reading to get the main take-aways.

High-content screening (HCS) in cancer research

In the first half of the webinar, Andrzej Mazan, Principal Investigator at Ryvu Therapeutics, described Ryvu’s ONCO Prime platform, a cutting-edge drug discovery pipeline for identifying novel synthetic lethal targets using patient-derived primary cell cultures. The platform facilitates high-throughput discovery of novel therapeutic targets with translational potential across various types of cancers. This integrated approach combines advanced disease models, CRISPR technology, and translational studies optimized for synthetic lethality research in oncology.

What distinguishes the ONCO Prime platform from other approaches is that it utilizes primary patient-derived cells and engineered models over traditional immortalized cell lines. This helps researchers avoid biases associated with cell lines (including issues such as genetic drift and lack of tumor representativity) and gain patient-specific insights into disease progression and treatment options. 

To develop patient-derived cell culture models, Ryvu has built a collaborative network through partnerships with various hospitals and universities in Poland to access primary patient samples. Additionally, they are utilizing integrated biobank and genomic databases for linking molecular and clinical data to enhance the predictive power of AI models.

Identifying synthetic lethality targets in patient-derived colorectal cancer (CRC) cell lines

The case study presented by Ryvu Therapeutics highlighted a recent project aimed at identifying novel, actionable therapeutic targets for colorectal cancer (CRC) treatments. This type of cancer is still associated with high mortality rates resulting from advanced-stage diagnoses and limited availability of targeted therapies for the majority of patients. The research sought to overcome this challenge by using models that better reflect the complexity of human tumors compared to traditional cell lines.

Ryvu scientists developed engineered intestinal epithelial stem cell models mimicking CRC evolution in combination with patient-derived samples. They conducted genome-wide CRISPR screenings to identify synthetic lethality targets selectively affecting mutant tumor cells without affecting healthy cells. The study successfully identified novel synthetic lethality targets not previously reported in public datasets and revealed potential synergistic drug combinations.

Using AI to enhance HCS workflows

In the second half of the webinar Krzysztof Rataj, a Chemoinformatics Data Scientist at Ardigen who specializes in using AI to enhance drug design and precision medicine, highlighted how deep learning and automated quality control (QC) workflows can significantly enhance efficiency, reduce costs, and improve data reliability in high-content screening for drug discovery.

Krzysztof explained the role Ardigen played in elucidating the synthetic lethality targets through HCS studies. AI and Machine Learning can significantly reduce the time and cost of HCS experiments by facilitating the analysis of DAPI-stained images using deep learning. In this case, a U-Net model was trained to segment cells and predict EdU staining specific for proliferating cancer cells based on DAPI (general nuclear staining) images.

The model achieved high accuracy in identifying proliferating cells from DAPI staining. The correlation between predicted areas of proliferating cells and EdU-stained cell counts reached a Pearson’s correlation of 0.99, enabling cost-efficient dose-response analysis and IC50 calculation with minimal training data (i.e., using a single plate stained with both DAPI and EdU). These reduced staining requirements in conjunction with AI-based analysis result in up to 500-fold cost reductions.

The importance of quality control (QC) in HCS data analysis

Krzysztof also highlighted the importance of ensuring data quality by identifying anomalies like poor cell distribution, mold contamination, and machine-related errors such as automatic dispensing malfunctions. Automatic QC procedures can be used to detect blurriness, dye spills, and channel intensity anomalies to ensure that the data from HCS studies can be trusted. Krzysztof used real experimental data to demonstrate how anomalous plates and wells were identified using Ardigen phenAID platform, even when conventional QC methods missed subtle issues.

In addition to standard QC measures, AI and ML tools can help mitigate the effects of challenging experimental conditions and implement batch corrections. Krzysztof highlighted advanced methods for identifying and correcting plate effects and maintaining uniformity across treatments and replicates. For example, you can leverage machine learning to assess wells with abnormal phenotypes. Ardigen’s QC methods allow parameter customization and offer tailored solutions for specific research needs. 

Implication of using advanced AI and HCS methods for drug discovery

The case studies presented by Ardigen and Ryvu demonstrate the value of advanced modeling platforms for discovering innovative cancer therapies, promising more effective treatment options for CRC and other tumor types. The presented methodologies are scalable and transferable to other applications as well.

HCS is a powerful method in itself and offers even more functionalities when used in combination with advanced AI and ML data analysis methods. Using AI for quality control improves the reliability of experimental results and helps identify potential issues impacting data quality. If you want to learn more about Ardigen’s AI-powered solutions for HCS, reach out to one of our experts.