Ex vivo discovery of synergistic drug combinations for hematologic malignancies.

Combination therapies have improved outcomes for patients with acute myeloid leukemia (AML). However, these patients still have poor overall survival. Although many combination therapies are identified with high-throughput screening (HTS), these approaches are constrained to disease models that can be grown in large volumes (e.g., immortalized cell lines), which have limited translational utility. To identify more effective and personalized treatments, we need better strategies for screening and exploring potential combination therapies. Our objective was to develop an HTS platform for identifying effective combination therapies with highly translatable ex vivo disease models that use size-limited, primary samples from patients with leukemia (AML and myelodysplastic syndrome). We developed a system, ComboFlow, that comprises three main components: MiniFlow, ComboPooler, and AutoGater. MiniFlow conducts ex vivo drug screening with a miniaturized flow-cytometry assay that uses minimal amounts of patient sample to maximize throughput. ComboPooler incorporates computational methods to design efficient screens of pooled drug combinations. AutoGater is an automated gating classifier for flow cytometry that uses machine learning to rapidly analyze the large datasets generated by the assay. We used ComboFlow to efficiently screen more than 3000 drug combinations across 20 patient samples using only 6 million cells per patient sample. In this screen, ComboFlow identified the known synergistic combination of bortezomib and panobinostat. ComboFlow also identified a novel drug combination, dactinomycin and fludarabine, that synergistically killed leukemic cells in 35 % of AML samples. This combination also had limited effects in normal, hematopoietic progenitors. In conclusion, ComboFlow enables exploration of massive landscapes of drug combinations that were previously inaccessible in ex vivo models. We envision that ComboFlow can be used to discover more effective and personalized combination therapies for cancers amenable to ex vivo models.

Predicting moderate drinking behaviors in National Health and Nutrition Examination Survey participants using biochemical and demographical factors with machine learning.

Recent studies revealed that any amount of alcohol consumption is an overall health detriment to multiple populations, contrary to popular beliefs. In addition, very few alcohol use studies utilized machine learning methods to compare the biological health of moderate drinkers compared to those that abstain from alcohol consumption, opting instead to focus on binge drinking and heavy drinking. Using participant data of multiple factor types from the National Health and Nutrition Examination Survey, we created prediction models with stacked ensembles and gradient boosting models. Machine learning models were used to identify which factors most enabled the prediction of moderate drinking behaviors. Our combined factor runs produced a cross-validation area under the curve (AUC) of 0.929 and a validation area under the curve of 0.806. Runs that only included biochemical or demographical factors received cross-validation AUC values of 0.825 and 0.925, and validation AUC values of 0.757 and 0.783, respectively. The top predictive factors for our machine learning runs, including gamma glutamyl transferase, gender, iron levels, and cigarette and marijuana usage, corroborate past studies that link those factors to alcohol consumption. Our findings identified key differences in the biological health of moderate drinkers compared to those that abstain from drinking. These results reveal a need to further explore the health effects of moderate drinking, especially for vulnerable populations.

Evaluation of a genetic risk score computed using human chromosomal-scale length variation to predict breast cancer.

INTRODUCTION: The ability to accurately predict whether a woman will develop breast cancer later in her life, should reduce the number of breast cancer deaths. Different predictive models exist for breast cancer based on family history, BRCA status, and SNP analysis. The best of these models has an accuracy (area under the receiver operating characteristic curve, AUC) of about 0.65. We have developed computational methods to characterize a genome by a small set of numbers that represent the length of segments of the chromosomes, called chromosomal-scale length variation (CSLV). METHODS: We built machine learning models to differentiate between women who had breast cancer and women who did not based on their CSLV characterization. We applied this procedure to two different datasets: the UK Biobank (1534 women with breast cancer and 4391 women who did not) and the Cancer Genome Atlas (TCGA) 874 with breast cancer and 3381 without. RESULTS: We found a machine learning model that could predict breast cancer with an AUC of 0.836 95% CI (0.830.0.843) in the UK Biobank data. Using a similar approach with the TCGA data, we obtained a model with an AUC of 0.704 95% CI (0.702, 0.706). Variable importance analysis indicated that no single chromosomal region was responsible for significant fraction of the model results. CONCLUSION: In this retrospective study, chromosomal-scale length variation could effectively predict whether or not a woman enrolled in the UK Biobank study developed breast cancer.

Development of a Machine Learning Model to Predict Recurrence of Oral Tongue Squamous Cell Carcinoma.

Despite diagnostic advancements, the development of reliable prognostic systems for assessing the risk of cancer recurrence still remains a challenge. In this study, we developed a novel framework to generate highly representative machine-learning prediction models for oral tongue squamous cell carcinoma (OTSCC) cancer recurrence. We identified cases of 5- and 10-year OTSCC recurrence from the SEER database. Four classification models were trained using the H2O ai platform, whose performances were assessed according to their accuracy, recall, precision, and the area under the curve (AUC) of their receiver operating characteristic (ROC) curves. By evaluating Shapley additive explanation contribution plots, feature importance was studied. Of the 130,979 patients studied, 36,042 (27.5%) were female, and the mean (SD) age was 58.2 (13.7) years. The Gradient Boosting Machine model performed the best, achieving 81.8% accuracy and 97.7% precision for 5-year prediction. Moreover, 10-year predictions demonstrated 80.0% accuracy and 94.0% precision. The number of prior tumors, patient age, the site of cancer recurrence, and tumor histology were the most significant predictors. The implementation of our novel SEER framework enabled the successful identification of patients with OTSCC recurrence, with which highly accurate and sensitive prediction models were generated. Thus, we demonstrate our framework's potential for application in various cancers to build generalizable screening tools to predict tumor recurrence.

Genetic Risk Scores and Missing Heritability in Ovarian Cancer.

Ovarian cancers are curable by surgical resection when discovered early. Unfortunately, most ovarian cancers are diagnosed in the later stages. One strategy to identify early ovarian tumors is to screen women who have the highest risk. This opinion article summarizes the accuracy of different methods used to assess the risk of developing ovarian cancer, including family history, BRCA genetic tests, and polygenic risk scores. The accuracy of these is compared to the maximum theoretical accuracy, revealing a substantial gap. We suggest that this gap, or missing heritability, could be caused by epistatic interactions between genes. An alternative approach to computing genetic risk scores, using chromosomal-scale length variation should incorporate epistatic interactions. Future research in this area should focus on this and other alternative methods of characterizing genomes.

A genetic risk score using human chromosomal-scale length variation can predict schizophrenia.

Studies indicate that schizophrenia has a genetic component, however it cannot be isolated to a single gene. We aimed to determine how well one could predict that a person will develop schizophrenia based on their germ line genetics. We compared 1129 people from the UK Biobank dataset who had a diagnosis of schizophrenia to an equal number of age matched people drawn from the general UK Biobank population. For each person, we constructed a profile consisting of numbers. Each number characterized the length of segments of chromosomes. We tested several machine learning algorithms to determine which was most effective in predicting schizophrenia and if any improvement in prediction occurs by breaking the chromosomes into smaller chunks. We found that the stacked ensemble, performed best with an area under the receiver operating characteristic curve (AUC) of 0.545 (95% CI 0.539-0.550). We noted an increase in the AUC by breaking the chromosomes into smaller chunks for analysis. Using SHAP values, we identified the X chromosome as the most important contributor to the predictive model. We conclude that germ line chromosomal scale length variation data could provide an effective genetic risk score for schizophrenia which performs better than chance.

A genetic risk score for glioblastoma multiforme based on copy number variations.

Glioblastoma multiforme is the most common form of brain cancer. Several lines of evidence suggest that glioblastoma multiforme has a genetic basis. A genetic test that could identify people who are at high risk of developing glioblastoma multiforme could improve our understanding of this form of brain cancer. Using the Cancer Genome Atlas (TCGA) dataset, we found common germ line DNA copy number variations in the TCGA population. We tested whether different sets of these germ line DNA copy number variations could effectively distinguish patients with glioblastoma multiforme from others in the TCGA dataset. We used a gradient boosting machine, a machine learning classification algorithm, to classify TCGA patients solely based on a set of germline DNA copy number variations. We found that this machine learning algorithm could classify TCGA glioblastoma multiforme patients from the other TCGA patients with an area under the curve (AUC) of the receiver operating characteristic curve (AUC=0.875). Grouped into quintiles, the highest ranked quintile by the machine learning algorithm had an odds ratio of 3.78 (95% CI 3.25-4.40) higher than the average odds ratio and about 40 (95% CI 20-70) times higher than the lowest quintile. The identification of an effective germ line genetic test to stratify risk of developing glioblastoma multiforme should lead to a better understanding of how this cancer forms. This result might ultimately lead to better treatments of glioblastoma multiforme.

Genetic risk score for ovarian cancer based on chromosomal-scale length variation.

INTRODUCTION: Twin studies indicate that a substantial fraction of ovarian cancers should be predictable from genetic testing. Genetic risk scores can stratify women into different classes of risk. Higher risk women can be treated or screened for ovarian cancer, which should reduce ovarian cancer death rates. However, current ovarian cancer genetic risk scores do not work that well. We developed a genetic risk score based on variations in the length of chromosomes. METHODS: We evaluated this genetic risk score using data collected by The Cancer Genome Atlas. We synthesized a dataset of 414 women who had ovarian serous carcinoma and 4225 women who had no form of ovarian cancer. We characterized each woman by 22 numbers, representing the length of each chromosome in their germ line DNA. We used a gradient boosting machine to build a classifier that can predict whether a woman had been diagnosed with ovarian cancer. RESULTS: The genetic risk score based on chromosomal-scale length variation could stratify women such that the highest 20% had a 160x risk (95% confidence interval 50x-450x) compared to the lowest 20%. The genetic risk score we developed had an area under the curve of the receiver operating characteristic curve of 0.88 (95% confidence interval 0.86-0.91). CONCLUSION: A genetic risk score based on chromosomal-scale length variation of germ line DNA provides an effective means of predicting whether or not a woman will develop ovarian cancer.

Evaluation of a genetic risk score for severity of COVID-19 using human chromosomal-scale length variation.

INTRODUCTION: The course of COVID-19 varies from asymptomatic to severe in patients. The basis for this range in symptoms is unknown. One possibility is that genetic variation is partly responsible for the highly variable response. We evaluated how well a genetic risk score based on chromosomal-scale length variation and machine learning classification algorithms could predict severity of response to SARS-CoV-2 infection. METHODS: We compared 981 patients from the UK Biobank dataset who had a severe reaction to SARS-CoV-2 infection before 27 April 2020 to a similar number of age-matched patients drawn for the general UK Biobank population. For each patient, we built a profile of 88 numbers characterizing the chromosomal-scale length variability of their germ line DNA. Each number represented one quarter of the 22 autosomes. We used the machine learning algorithm XGBoost to build a classifier that could predict whether a person would have a severe reaction to COVID-19 based only on their 88-number classification. RESULTS: We found that the XGBoost classifier could differentiate between the two classes at a significant level (p = 2 · 10-11) as measured against a randomized control and (p = 3 · 10-14) as measured against the expected value of a random guessing algorithm (AUC = 0.5). However, we found that the AUC of the classifier was only 0.51, too low for a clinically useful test. CONCLUSION: Genetics play a role in the severity of COVID-19, but we cannot yet develop a useful genetic test to predict severity.

Laboratory tests as short-term correlates of stroke.

BACKGROUND: The widespread adoption of electronic health records provides new opportunities to better predict which patients are likely to suffer a stroke. Using electronic health records, we assessed the correlation of different laboratory tests to future occurrences of a stroke. METHODS: We examined the electronic health records of 2.4 million people over a two year time span. These records contained 26,964 diagnoses of stroke. Using Cox regression analysis, we measured whether any one of 1796 different laboratory tests were effectively correlated with a future diagnosis of stroke. RESULTS: We identified 38 different laboratory tests that had significant short-term (two year) prognostic value for a future diagnosis of stroke. For each of the 38 laboratory tests we also compiled the Kaplan-Meier survival curve, and relative risk ratio that the test confers. CONCLUSION: Several dozen laboratory tests are effective short-term correlates of stroke.

Breast tumor laterality in the United States depends upon the country of birth, but not race.

More breast cancers are diagnosed in the left breast than the right. The ratio (l/r) is called the laterality ratio. We analyzed 1.2 million cases of breast cancer diagnosed in the US between 1973 and 2010 and recorded by the Surveillance, Epidemiology, and End Results (SEER) program. We found that the laterality ratio depends upon the country of birth, but not race of the patient. We identified five countries of birth that had p-values larger than 0.995, while we expected to see less than 1. Those born in Japan (l/r = 1.14, p = 0.997), the Ryukyu Islands (l/r = 2.6, p = 0.998), Laos (l/r = 1.62, p = 0.9999) and Algeria (l/r = 2.1 p = 0.9959) had significantly larger laterality ratios compared to the overall SEER population (l/r = 1.04). Those born in Poland (l/r = 0.92, p = 0.997) had a laterality ratio significantly less than expected. We compared the laterality ratio calculated for tumors occurring in each quadrant of the breast for two immigrant populations: those born in Japan and those born in Poland. We found the only significant difference was in the laterality ratio of the upper outer quadrant. Thus, the birthplace effect appears to only occur in the upper outer quadrant of the breast. Finally, we found a small, but statistically significant, increase in the breast cancer laterality ratio with age, and decrease with birth year and year of diagnosis.

Similarities in the Age-Specific Incidence of Colon and Testicular Cancers.

Colon cancers are thought to be an inevitable result of aging, while testicular cancers are thought to develop in only a small fraction of men, beginning in utero. These models of carcinogenesis are, in part, based upon age-specific incidence data. The specific incidence for colon cancer appears to monotonically increase with age, while that of testicular cancer increases to a maximum value at about 35 years of age, then declines to nearly zero by the age of 80. We hypothesized that the age-specific incidence for these two cancers is similar; the apparent difference is caused by a longer development time for colon cancer and the lack of age-specific incidence data for people over 84 years of age. Here we show that a single distribution can describe the age-specific incidence of both colon carcinoma and testicular cancer. Furthermore, this distribution predicts that the specific incidence of colon cancer should reach a maximum at about age 90 and then decrease. Data on the incidence of colon carcinoma for women aged 85-99, acquired from SEER and the US Census, is consistent with this prediction. We conclude that the age specific data for testicular cancers and colon cancers is similar, suggesting that the underlying process leading to the development of these two forms of cancer may be similar.

Age-specific incidence data indicate four mutations are required for human testicular cancers.

Normal human cells require a series of genetic alterations to undergo malignant transformation. Direct sequencing of human tumors has identified hundreds of mutations in tumors, but many of these are thought to be unnecessary and a result of, rather than a cause of, the tumor. The exact number of mutations to transform a normal human cell into a tumor cell is unknown. Here I show that male gonadal germ cell tumors, the most common form of testicular cancers, occur after four mutations. I infer this by constructing a mathematical model based upon the multi-hit hypothesis and comparing it to the age-specific incidence data. This result is consistent with the multi-hit hypothesis, and implies that these cancers are genetically or epigenetically predetermined at birth or an early age.

Reverse engineering the yeast RNR1 transcriptional control system.

Transcription is controlled by multi-protein complexes binding to short non-coding regions of genomic DNA. These complexes interact combinatorially. A major goal of modern biology is to provide simple models that predict this complex behavior. The yeast gene RNR1 is transcribed periodically during the cell cycle. Here, we present a pilot study to demonstrate a new method of deciphering the logic behind transcriptional regulation. We took regular samples from cell cycle synchronized cultures of Saccharomyces cerevisiae and extracted nuclear protein. We tested these samples to measure the amount of protein that bound to seven different 16 base pair sequences of DNA that have been previously identified as protein binding locations in the promoter of the RNR1 gene. These tests were performed using surface plasmon resonance. We found that the surface plasmon resonance signals showed significant variation throughout the cell cycle. We correlated the protein binding data with previously published mRNA expression data and interpreted this to show that transcription requires protein bound to a particular site and either five different sites or one additional sites. We conclude that this demonstrates the feasibility of this approach to decipher the combinatorial logic of transcription.

Parallel routes of human carcinoma development: implications of the age-specific incidence data.

BACKGROUND: The multi-stage hypothesis suggests that cancers develop through a single defined series of genetic alterations. This hypothesis was first suggested over 50 years ago based upon age-specific incidence data. However, recent molecular studies of tumors indicate that multiple routes exist to the formation of cancer, not a single route. This parallel route hypothesis has not been tested with age-specific incidence data. METHODOLOGY/PRINCIPAL FINDINGS: To test the parallel route hypothesis, I formulated it in terms of a mathematical equation and then tested whether this equation was consistent with age-specific incidence data compiled by the Surveillance Epidemiology and End Results (SEER) cancer registries since 1973. I used the chi-squared goodness of fit test to measure consistency. The age-specific incidence data from most human carcinomas, including those of the colon, lung, prostate, and breast were consistent with the parallel route hypothesis. However, this hypothesis is only consistent if an immune sub-population exists, one that will never develop carcinoma. Furthermore, breast carcinoma has two distinct forms of the disease, and one of these occurs at significantly different rates in different racial groups. CONCLUSIONS/SIGNIFICANCE: I conclude that the parallel route hypothesis is consistent with the age-specific incidence data only if carcinoma occurs in a distinct sub population, while the multi-stage hypothesis is inconsistent with this data.

Microfabricated arrays of cylindrical wells facilitate single-molecule enzymology of alpha-chymotrypsin.

Single-molecule enzymology allows scientists to examine the distributions of kinetic rates among members of a population. We describe a simple method for the analysis of single-molecule enzymatic kinetics and provide comparisons to ensemble-averaged kinetics. To isolate our model enzyme, alpha-chymotrypsin, into single molecules, we use an array of cylindrical poly(dimethylsiloxane) wells 2 microm in diameter and 1.35 microm in height. Inside the wells, a protease assay with a profluorescent substrate detects alpha-chymotrypsin activity. We hold the concentration of alpha-chymotrypsin at 0.39 nM in a given well with an enzyme-to-substrate ratio of 1:6,666 molecules. Fluorescence emitted by the substrate is proportional to enzyme activity and detectable by a charge-coupled device. This method allows for the simultaneous real-time characterization of hundreds of individual enzymes. We analyze single-molecule kinetics by recording and observing their intensity trajectories over time. By testing our method with our current instruments, we confirm that our methodology is useful for the analysis of single enzymes for extracting static inhomogeneity.

Dynamic SPR monitoring of yeast nuclear protein binding to a cis-regulatory element.

Gene expression is controlled by protein complexes binding to short specific sequences of DNA, called cis-regulatory elements. Expression of most eukaryotic genes is controlled by dozens of these elements. Comprehensive identification and monitoring of these elements is a major goal of genomics. In pursuit of this goal, we are developing a surface plasmon resonance (SPR) based assay to identify and monitor cis-regulatory elements. To test whether we could reliably monitor protein binding to a regulatory element, we immobilized a 16bp region of Saccharomyces cerevisiae chromosome 5 onto a gold surface. This 16bp region of DNA is known to bind several proteins and thought to control expression of the gene RNR1, which varies through the cell cycle. We synchronized yeast cell cultures, and then sampled these cultures at a regular interval. These samples were processed to purify nuclear lysate, which was then exposed to the sensor. We found that nuclear protein binds this particular element of DNA at a significantly higher rate (as compared to unsynchronized cells) during G1 phase. Other time points show levels of DNA-nuclear protein binding similar to the unsynchronized control. We also measured the apparent association complex of the binding to be 0.014s(-1). We conclude that (1) SPR-based assays can monitor DNA-nuclear protein binding and that (2) for this particular cis-regulatory element, maximum DNA-nuclear protein binding occurs during G1 phase.

Diversity in the activity of individual enzymes.

Although the structure of an enzyme is often depicted as static, it is dynamic. Hence, a population of chemically identical enzymes has not one, but a distribution of structures at any moment in time. Does this have an effect on the activity of the enzyme? This article reviews experiments designed to test the hypothesis that this distribution of structures results in a distribution of enzyme activities. The experiments reviewed here use different enzymes, falvin adenine dinucleotide, beta-galactosidase, alkaline phosphatase, exonuclease I, lactate dehydrogenase I, alpha-chymotrypsin, the 20S proteasome, and horseradish peroxidase. All experiments come to the same conclusion, when measured individually, apparently identical enzymes show a distribution in rates of activity.

Single-molecule enzymology of chymotrypsin using water-in-oil emulsion.

Single-molecule studies allow the study of subtle activity differences due to local folding in proteins, but are time consuming and difficult because only a few molecules are observed in one experiment. We developed an assay where we can simultaneously measure the activity of hundreds of individual molecules. The assay utilizes a synthetic chymotrypsin substrate that is nonfluorescent before cleavage by chymotrypsin, but is intensely fluorescent afterward. We encapsulated the enzyme and substrate in micron-sized droplets of water surrounded by silicone oil where each microdroplet contains <1 enzyme on average. A microscope and charge-coupled device camera are used to measure the fluorescence intensity of the same individual droplet over time. Based on these measurements, we conclude that enzymatic reactions could occur within this emulsion system, the statistical average activity of individual chymotrypsin molecules is similar to that measured in bulk, and the activity of individual chymotrypsin is heterogeneous.