April 20, 2024
Disease Treatments

Using Transfer Learning to Revolutionize Disease Treatments

Advancements in gene sequencing and computing technology have paved the way for a surge in bioinformatic data availability and processing power. This has created an optimal environment for artificial intelligence (AI) to drive strategies aimed at manipulating cell behavior for new disease treatments. Researchers at Northwestern University have leveraged this technological intersection by introducing an AI-powered transfer learning methodology that repurposes existing data to forecast gene perturbation combinations capable of transforming cell types or revitalizing diseased cells.

The study titled ‘Cell Reprogramming Design by Transfer Learning of Functional Transcriptional Networks’ has been recently published in the Proceedings of the National Academy of Sciences.

Following the completion of the human genome project two decades ago, scientists have grappled with understanding how over 20,000 genes collaborate to govern the multitude of cell types in the human body. Through systematic trial-and-error methods, researchers have demonstrated the feasibility of reprogramming cell types by manipulating a small set of genes. Advancements in sequencing technologies post the human genome project not only made genetic code reading more affordable but also enabled gene expression measurement, quantifying protein precursors crucial for cell functions.

The drop in sequencing costs led to a massive repository of publicly available bioinformatics data, opening avenues for synthesizing this data to design gene manipulations that prompt desired cell behaviors. The ability to steer cell behavior for transitioning across various cell types can find applications in regenerating injured tissues or converting cancer cells into normal cells.

Injuries such as strokes, arthritis, and multiple sclerosis afflict around 2.9 million individuals annually in the United States, costing up to $400 million yearly. On the other hand, cancers lead to nearly 10 million deaths globally every year, with economic costs running into trillions of dollars. The inadequacy of current standard treatments to regenerate tissues or offer comprehensive efficacy underlines the urgency for developing more potent, widely applicable therapies necessitating the identification of molecular interventions inferred from high-throughput data.

In the new study, the researchers train an AI model to comprehend how gene expression governs cell behavior using publicly available gene expression data. The model derived from this learning process is then utilized for specific cell reprogramming purposes, identifying the gene manipulations most likely to trigger desired cell type transitions.

Thomas Wytock, the lead author of the paper and a member of the Center for Network Dynamics at Northwestern University, stated, “Our work distinguishes itself from prior approaches in rationally designing strategies to manipulate cell behavior.” Traditionally, such approaches categorized genes into networks based on their interactions or properties, or compared gene expressions in healthy and diseased cells to pinpoint significant differences.

Northwestern’s novel methodology amalgamates the strengths of both types of models, encompassing all cell genes quantitatively portraying their expressions. This is achieved by condensing the expressions of nearly 20,000 genes into 10 linear combinations termed eigengenes, simplifying the vast dynamical network dynamics to a few essential components.

The newfound method overcomes the challenge of understanding how genes act together to induce cell behavior changes by acknowledging that genes revolutionize their expressions collectively. This property is quantitatively addressed in terms of eigengenes, enabling the additive combination of their responses to different gene perturbations. Equipped with this AI model, the researchers collated publicly available data to foresee gene combinations likely to induce desired reprogramming transitions, for instance, converting diseased cells into healthy ones.

The approach’s efficacy lies in its computational ability to assess numerous combinations, allowing for predictive comparisons across a multitude of scenarios. Furthermore, the method’s adaptability to consider gene expressions additively promotes its generalization across cell types, streamlining the process of curing diseases involving cellular dysfunction such as cancers, diabetes, and autoimmune diseases.

The AI-powered approach, viewed as a platform, can accommodate disease-specific genetic data for individual patients. Contextualizing gene expression data across the vast archives of the National Center for Biotechnology Information holds promise for devising precise predictions on how genes collaborate to regulate normal and diseased cell behaviors.

Note:
1. Source: Coherent Market Insights, Public sources, Desk research.
2. We have leveraged AI tools to mine information and compile it.