Technological advances in gene sequencing and computing have led to an explosion in the availability of bioinformatic data and processing power, respectively, creating a ripe nexus to design strategies for controlling cell behavior.

OncoDxRx scientists have reaped fruit from this nexus by developing a liquid biopsy-powered gene expression approach, PGA (Patient-derived Gene expression-informed Anticancer drug efficacy), that captures cancer patient’s genetic fingerprint data to predict drug response, therapeutic outcome and recurrence risk.

Since the completion of the human genome project 20 years ago, scientists have known that human DNA comprises more than 20,000 genes. However, it has remained a mystery as to how these genes work together to orchestrate the hundreds of different cell types in our body.

Surprisingly, essentially by guided trial-and-error, researchers have demonstrated that it is possible to “reprogram” cell type by manipulating only a handful of genes. The human genome project also facilitated advances in sequencing technologies, making it cheaper not only to read the genetic code, but also to measure gene expression, which quantifies the precursors of the proteins that carry out cell functions.

This increase in affordability has led to the accumulation of a massive amount of publicly available bioinformatic data, raising the possibility of synthesizing these data to rationally design gene panels that can be powerfully predictive of cell behaviors. The ability to forecast cell behavior, and thus transitions across cell types, can be applied to predicting cancer cell susceptibility to anticancer drugs.

Meanwhile, cancers are responsible for around 10 million deaths annually worldwide with economic costs in the trillions of dollars.

Because the current standard of care does not regenerate tissues and/or has limited efficacy, there is a critical need to develop more effective treatments that are broadly applicable, which in turn requires identification of molecular interventions that can be inferred from high-throughput data.

OncoDxRx’s team performs gene expression (transcriptomic) profiling to learn how gene expression gives rise to tumor behavior using liquid biopsy cell-free mRNA (cfmRNA) from patient’s blood. The specific gene expression signature generated by this process is then transferred to in silico data fusion analytics. PGA technology predicts and selects top-ranking drugs that are most likely to induce positive response and benefit to the patient.

Unprecedented exploration of the gene activity dynamics

Gene expression approaches mostly fall into two categories: one in which genes are organized into networks according to their interactions or common properties; and another in which the expression of genes from healthy and diseased cells are compared to single out the genes that show the largest differences.”

In the first category, there is a tradeoff between realism and scale. Some gene panels comprise many genes but only say whether a relationship is present or absent. Other panels are quantitative and experimentally validated but necessarily involve a small number of genes and relationships.

OncoDxRx’s PGA retains the strengths of both types of models: it is inclusive of critical genes in the cell and quantitative in representing their expressions. This is achieved by reducing the expression of nearly 20,000 individual genes to no more than a dozen of linear combinations of such genes, which are weighted averages referred to as eigengenes.

“Eigengenes basically show how genes operate in concert, making it possible to simplify the dynamics of a large dynamical network to just a few moving parts,” said OncoDxRx. “Each eigengene can be thought of as a generalized pathway that is approximately independent of the others. So, eigengenes pick up the relevant correlations and independences in the gene regulatory network.”

Approaches in the second category can find individual genes associated with a change in cell behavior but fail to specify how genes work together to enable this change. OncoDxRx’s PGA overcomes this challenge by recognizing that genes change their expressions in concert. The quantitative accounting of this property in terms of eigengenes makes it possible to additively combine their responses to different gene perturbations by suitably scaling them. The combined responses can then be input into the process to determine which drugs elicit the tumor response.

Averting combinatorial explosion

Equipped with PGA, the team curated publicly available databases to identify gene-drug correlation. They then developed a data fusion and transformation computation to predict gene expression patterns that are most likely to represent a desired cell phenotype, such as drug response. The approach that results from integrating the data and computation circumvents combinatorial explosion in order to identify the effective ones. This is significant because experiments can test only a limited number of cases, and PGA provides a way to identify the most promising drugs to be used in clinic.

“This is a case in which the whole is well approximated by the sum of the parts,” OncoDxRx said. “This property of the interventions needed to induce transitions between cell types is counterintuitive because the cell types themselves emerge from collective interactions among genes.”

Because PGA addresses the main gene-to-drug challenges, it can be applied to many different biomedical conditions, including those that will benefit from future data.

A flexible model for forthcoming data

The fact that responses to gene activities combine additively facilitates generalization across cell types. For example, if a gene is activated in a skin cell, the resulting impact on expression would be largely the same in a liver cell.

Thus, the PGA-powered approach can be thought of as a platform into which data pertaining to a specific disease in a specific patient may be inserted. The approach may be applied whenever curing the disease can be conceived as a gene malfunction problem, as in the case of cancers, which all result from gene expression dysregulation.

OncoDxRx’s work provides a critical tool for translating this wealth of data into specific predictions of how genes work together to control the behavior of normal and cancer cells.