A primer on applying AI synergistically with domain expertise to oncology.

BACKGROUND: The concurrent growth of large-scale oncology data alongside the computational methods with which to analyze and model it has created a promising environment for revolutionizing cancer diagnosis, treatment, prevention, and drug discovery. Computational methods applied to large datasets have accelerated the drug discovery process by reducing bottlenecks and widening the search space beyond what is experimentally tractable. As the research community gains understanding of the myriad genetic underpinnings of cancer via sequencing, imaging, screens, and more that are ingested, transformed, and modeled by top open-source machine learning and artificial intelligence tools readily available, the next big drug candidate might seem merely an "Enter" key away. Of course, the reality is more convoluted, but still promising. SCOPE OF REVIEW: We present methods to approach the process of building an AI model, with strong emphasis on the aspects of model development we believe to be crucial to success but that are not commonly discussed: diligence in posing questions, identifying suitable datasets and curating them, and collaborating closely with biology and oncology experts while designing and evaluating the model. Digital pathology, Electronic Health Records, and other data types outside of high-throughput molecular data are reviewed well by others and outside of the scope of this review. This review emphasizes the importance of considering the limitations of the datasets, computational methods, and our minds when designing AI models. For example, datasets can be biased towards areas of research interest, funding, and particular patient populations. Neural networks may learn representations and correlations within the data that are grounded not in biological phenomena, but statistical anomalies erroneously extracted from the training data. Researchers may mis-interpret or over-interpret the output, or design and evaluate the training process such that the resultant model generalizes poorly. Fortunately, awareness of the strengths and limitations of applying data analytics and AI to drug discovery enables us to leverage them carefully and insightfully while maximizing their utility. These applications when performed in close collaboration with domain experts, together with continuous critical evaluation, generation of new data to minimize known blind spots as they are found, and rigorous experimental validation, increases the success rate of the study. We will discuss applications including AI-assisted target identification, drug repurposing, patient stratification, and gene prioritization. MAJOR CONCLUSIONS: Data analytics and AI have demonstrated capabilities to revolutionize cancer research, prevention, and treatment by maximizing our understanding and use of the expanding panoply of experimental data. However, to separate promise from true utility, computational tools must be carefully designed, critically evaluated, and constantly improved. Once that is achieved, a human-computer hybrid discovery process will outperform one driven by each alone. GENERAL SIGNIFICANCE: This review highlights the challenges and promise of synergizing predictive AI models with human expertise towards greater understanding of cancer.

A systematic approach to biomarker discovery; preamble to "the iSBTc-FDA taskforce on immunotherapy biomarkers".

The International Society for the Biological Therapy of Cancer (iSBTc) has initiated in collaboration with the United States Food and Drug Administration (FDA) a programmatic look at innovative avenues for the identification of relevant parameters to assist clinical and basic scientists who study the natural course of host/tumor interactions or their response to immune manipulation. The task force has two primary goals: 1) identify best practices of standardized and validated immune monitoring procedures and assays to promote inter-trial comparisons and 2) develop strategies for the identification of novel biomarkers that may enhance our understating of principles governing human cancer immune biology and, consequently, implement their clinical application. Two working groups were created that will report the developed best practices at an NCI/FDA/iSBTc sponsored workshop tied to the annual meeting of the iSBTc to be held in Washington DC in the Fall of 2009. This foreword provides an overview of the task force and invites feedback from readers that might be incorporated in the discussions and in the final document.

Compositional differences between Copaxone and Glatopa are reflected in altered immunomodulation ex vivo in a mouse model.

Copaxone (glatiramer acetate, GA), a structurally and compositionally complex polypeptide nonbiological drug, is an effective treatment for multiple sclerosis, with a well-established favorable safety profile. The short antigenic polypeptide sequences comprising therapeutically active epitopes in GA cannot be deciphered with state-of-the-art methods; and GA has no measurable pharmacokinetic profile and no validated pharmacodynamic markers. The study reported herein describes the use of orthogonal standard and high-resolution physicochemical and biological tests to characterize GA and a U.S. Food and Drug Administration-approved generic version of GA, Glatopa (USA-FoGA). While similarities were observed with low-resolution or destructive tests, differences between GA and USA-FoGA were measured with high-resolution methods applied to an intact mixture, including variations in surface charge and a unique, high-molecular-weight, hydrophobic polypeptide population observed only in some USA-FoGA lots. Consistent with published reports that modifications in physicochemical attributes alter immune-related processes, genome-wide expression profiles of ex vivo activated splenocytes from mice immunized with either GA or USA-FoGA showed that 7-11% of modulated genes were differentially expressed and enriched for immune-related pathways. Thus, differences between USA-FoGA and GA may include variations in antigenic epitopes that differentially activate immune responses. We propose that the assays reported herein should be considered during the regulatory assessment process for nonbiological complex drugs such as GA.

Functional effects of the antigen glatiramer acetate are complex and tightly associated with its composition.

Glatiramer acetate (Copaxone®; GA) is a non-biological complex drug for multiple sclerosis. GA modulated thousands of genes in genome-wide expression studies conducted in THP-1 cells and mouse splenocytes. Comparing GA with differently-manufactured glatiramoid Polimunol (Synthon) in mice yielded hundreds of differentially expressed probesets, including biologically-relevant genes (e.g. Il18, adj p<9e-6) and pathways. In human monocytes, 700+ probesets differed between Polimunol and GA, enriching for 130+ pathways including response to lipopolysaccharide (adj. p<0.006). Key differences were confirmed by qRT-PCR (splenocytes) or proteomics (THP-1). These studies demonstrate the complexity of GA's mechanisms of action, and may help inform therapeutic equivalence assessment.

Leveraging existing data sets to generate new insights into Alzheimer's disease biology in specific patient subsets.

To generate new insights into the biology of Alzheimer's Disease (AD), we developed methods to combine and reuse a wide variety of existing data sets in new ways. We first identified genes consistently associated with AD in each of four separate expression studies, and confirmed this result using a fifth study. We next developed algorithms to search hundreds of thousands of Gene Expression Omnibus (GEO) data sets, identifying a link between an AD-associated gene (NEUROD6) and gender. We therefore stratified patients by gender along with APOE4 status, and analyzed multiple SNP data sets to identify variants associated with AD. SNPs in either the region of NEUROD6 or SNAP25 were significantly associated with AD, in APOE4+ females and APOE4+ males, respectively. We developed algorithms to search Connectivity Map (CMAP) data for medicines that modulate AD-associated genes, identifying hypotheses that warrant further investigation for treating specific AD patient subsets. In contrast to other methods, this approach focused on integrating multiple gene expression datasets across platforms in order to achieve a robust intersection of disease-affected genes, and then leveraging these results in combination with genetic studies in order to prioritize potential genes for targeted therapy.

Gene expression studies of a human monocyte cell line identify dissimilarities between differently manufactured glatiramoids.

Glatiramer Acetate (GA) has provided safe and effective treatment for multiple sclerosis (MS) patients for two decades. It acts as an antigen, yet the precise mechanism of action remains to be fully elucidated, and no validated pharmacokinetic or pharmacodynamic biomarkers exist. In order to better characterize GA's biological impact, genome-wide expression studies were conducted with a human monocyte (THP-1) cell line. Consistent with previous literature, branded GA upregulated anti-inflammatory markers (e.g. IL10), and modulated multiple immune-related pathways. Despite some similarities, significant differences were observed between expression profiles induced by branded GA and Probioglat, a differently-manufactured glatiramoid purported to be a generic GA. Key results were verified using qRT-PCR. Genes (e.g. CCL5, adj. p < 4.1 × 10(-5)) critically involved in pro-inflammatory pathways (e.g. response to lipopolysaccharide, adj. p = 8.7 × 10(-4)) were significantly induced by Probioglat compared with branded GA. Key genes were also tested and confirmed at the protein level, and in primary human monocytes. These observations suggest differential biological impact by the two glatiramoids and warrant further investigation.

Comparing the biological impact of glatiramer acetate with the biological impact of a generic.

For decades, policies regarding generic medicines have sought to provide patients with economical access to safe and effective drugs, while encouraging the development of new therapies. This balance is becoming more challenging for physicians and regulators as biologics and non-biological complex drugs (NBCDs) such as glatiramer acetate demonstrate remarkable efficacy, because generics for these medicines are more difficult to assess. We sought to develop computational methods that use transcriptional profiles to compare branded medicines to generics, robustly characterizing differences in biological impact. We combined multiple computational methods to determine whether differentially expressed genes result from random variation, or point to consistent differences in biological impact of the generic compared to the branded medicine. We applied these methods to analyze gene expression data from mouse splenocytes exposed to either branded glatiramer acetate or a generic. The computational methods identified extensive evidence that branded glatiramer acetate has a more consistent biological impact across batches than the generic, and has a distinct impact on regulatory T cells and myeloid lineage cells. In summary, we developed a computational pipeline that integrates multiple methods to compare two medicines in an innovative way. This pipeline, and the specific findings distinguishing branded glatiramer acetate from a generic, can help physicians and regulators take appropriate steps to ensure safety and efficacy.

Nucleic acid and protein mass mapping by live-cell deep-ultraviolet microscopy.

We developed a deep-ultraviolet (UV) microscope capable of imaging cell mitosis and motility at 280 nm for 45 min with minimal UV-induced toxicity, and for 6 h before the onset of visible cell death in cultured human and mouse cells. Combined with computational methods that convert the intensity of each pixel into an estimate of mass, deep-UV microscopy images generate maps of nucleic acid mass, protein mass and fluorescence yield in unlabeled cells.