NEW YORK – Immunis.AI's immunogenomic Intelligentia platform accurately detected early-stage prostate cancer and estimated an individual's risk for that cancer growing aggressive in a recently published study.
The findings, published in the journal Cells, demonstrate the platform's potential to detect early-stage prostate cancer and to identify which patients may need active surveillance based on a prostate cancer risk assessment. The findings also show the potential for the Intelligentia platform to assist physicians in patient screening, in treatment decisions, and in detecting minimal residual disease.
Christopher Thibodeau, CEO of Immunis.AI, called these results "a step in the right direction" toward validating the Intelligentia platform as a lab-developed test. The company hopes to launch in late 2022.
The study consisted of 1,018 men either known to have untreated prostate cancer or suspected of having prostate cancer and sought to determine if the differential expression of CD2 positive and CD14 positive cells associated with features of more aggressive early-stage disease, generally while the cancer remains clinically localized. Study participants were followed for a median of 3.8 years.
The transition from localized to aggressive cancer marks an escape point from the immune system and accompanies a sharp change in the expression profile of prostate cancer-related cells such as CD2+ and CD14+ cells.
The assay, combined with traditional clinical risk factors such as age, serum prostate-specific antigen, PSA density, race, digital rectal examination, and family history, assigned risk of aggressive prostate cancer with high sensitivity and low false positive rate.
The core methodology consists of isolating and comparing different subsets of white blood cells with different mechanisms of action (phagocytic and immune response) within a single patient, effectively using that patient's own cells as controls.
"Using the patient's one cell type against another cell type as their own control really helps us reduce the noise and find signals that are otherwise difficult to find," explained Kirk Wojno, chief medical officer of Immunis.AI.
Transcription profiles of each cell, obtained via whole transcriptome RNA-seq and currently using an Illumina HiSeq 2500, are then evaluated separately, while normalized transcript counts are fed into the model. Normalization consists of subtracting log-transformed counts from one cell type from the log-transformed counts of the comparator cell type for each patient, on a gene-by-gene basis. In the published study, for instance, CD14-related counts were normalized to CD2-related counts by subtracting log-transformed CD2 counts from the log-transformed CD14 counts.
Transcripts are down-selected in an unsupervised, AI-assisted manner to avoid overfitting and then sent, along with other clinical parameters such as age and PSA levels, as inputs into the model, which outputs patient risk predictions.
Wojno explained that the subtraction method helps reduce the number of genes in a dataset to the most informative ones.
"We then optimize that to find those with the highest signal and the least noise," he added. "This is very personalized medicine compared to what everybody else is doing out in this space."
The subtraction method also sits at the center of Immunis.AI's patenting strategy.
"The core patents on the technology are method-based patents that would protect that subtraction," Geoffrey Erickson, senior VP and cofounder of Immunis.AI, said in an interview. "In addition to that, the test would be protected by the series of [transcription] signatures that arise from the use of that method."
Each signature forms a set of biomarkers meant to help clinicians make decisions such as determining which patients should be on active surveillance and which should not. By observing changes in patient immune cells, the platform goes beyond disease detection and surveillance to also enable monitoring of patient immune responses to disease.
The published study is one of three prospective studies aimed at demonstrating Intelligentia's capacity for early cancer detection and assessment of disease aggressiveness. Whereas the currently published study was conducted under very stringent conditions in order to develop the platform, the other two studies make use of more real-world conditions and an optimized collection technique developed to accommodate a more commercially friendly format for testing in the company's centralized lab. Both studies have completed enrollment and Immunis.AI expects to publish the results of the second study in the coming months and those of the third study shortly thereafter.
Although prostate cancer detection is the first indication to which the LDT will be applied, the company is working toward expanding it in the future as a multi-cancer test and as a pan-disease assay.
Immunis.AI is currently in the process of designing a study with multiple academic sites, focused on developing fit-for-purpose signatures for breast, colon, and lung cancers.
Thibodeau pointed out that assessing total transcriptomics through RNA-seq broadens the platform's utility, enabling researchers and clinicians to ask multiple different questions using a single patient blood sample.
"It is a universal application in that we're taking a sample and doing total transcriptomics on it — bulk sequencing using the Illumina platform," he said. "From there, we apply machine learning to tease out the genes of interest for a given disease."
The platform's capabilities won't apply to every disease going forward, Thibodeau explained, "but from that dataset that we generate, which is universal, we can interrogate it for different clinical questions."
Immunis.AI plans to launch its platform toward the end of 2022. Launch was initially slated for later this year but had to be postponed due to the global COVID-19 pandemic.