Molecular and translational biology of the blood-based VeriStrat® proteomic test used in cancer immunotherapy treatment guidance.

Introduction: The VeriStrat® test (VS) is a blood-based assay that predicts a patient's response to therapy by analyzing eight features in a spectrum obtained from matrix-assisted laser desorption/ionization time-of-flight (MALDI-TOF) analysis of human serum and plasma. In a recent analysis of the INSIGHT clinical trial (NCT03289780), it was found that the VS labels, VS Good and VS Poor, can effectively predict the responsiveness of non-small cell lung cancer (NSCLC) patients to immune checkpoint inhibitor (ICI) therapy. However, while VS measures the intensities of spectral features using MALDI-TOF analysis, the specific proteoforms underlying these features have not been comprehensively identified. Objectives: The objective of this study was to identify the proteoforms that are measured by VS. Methods: To resolve the features obtained from the low-resolution MALDI-TOF procedure used to acquire mass spectra for VS DeepMALDI® analysis of serum was employed. This technique allowed for the identification of finer peaks within these features. Additionally, a combination of reversed-phase fractionation and liquid chromatography-tandem mass spectrometry (LC-MS/MS) was then used to identify the proteoforms associated with these peaks. Results: The analysis revealed that the primary constituents of the spectrum measured by VS are serum amyloid A1, serum amyloid A2, serum amyloid A4, C-reactive protein, and beta-2 microglobulin. Conclusion: Proteoforms involved in host immunity were identified as significant components of these features. This newly acquired information improves our understanding of how VS can accurately predict patient response to therapy. It opens up additional studies that can expand our understanding even further.

Folding of Staphylococcal Nuclease Induced by Binding of Chemically Modified Substrate Analogues Sheds Light on Mechanisms of Coupled Folding/Binding Reactions.

Several proteins have been shown to undergo a shift in the mechanism of ligand binding-induced folding from conformational selection (CS; folding precedes binding) to induced fit (IF; binding precedes folding) with increasing ligand concentration. In previous studies of the coupled folding/binding reaction of staphylococcal nuclease (SNase) in the presence of a substrate analogue, adenosine-3',5'-diphosphate (prAp), we found that the two phosphate groups make important energetic contributions toward stabilizing its complex with the native protein as well as transient conformational states encountered at high ligand concentrations favoring IF. However, the structural contributions of each phosphate group during the reaction remain unclear. To address this question, we relied on fluorescence, nuclear magnetic resonance (NMR), absorption, and isothermal titration calorimetry to study the effects of deletion of the phosphate groups of prAp on the kinetics of ligand-induced folding, using a strategy analogous to mutational ϕ-value analysis to interpret the results. Kinetic measurements over a wide range of ligand concentrations, together with structural characterization of a transient protein-ligand encounter complex using 2D NMR, indicated that, at high ligand concentrations favoring IF, (i) the 5'-phosphate group interacts weakly with denatured SNase during early stages of the reaction, resulting in loose docking of the two domains of SNase, and (ii) the 3'-phosphate group engages in some specific contacts with the polypeptide in the transition state prior to formation of the native SNase-prAp complex.

PDLIM3 supports hedgehog signaling in medulloblastoma by facilitating cilia formation.

Elevated levels of PDLIM3 expression are frequently detected in sonic hedgehog (SHH) group of medulloblastoma (MB). However, the possible role of PDLIM3 in MB tumorigenesis is still unknown. Here, we found that PDLIM3 expression is necessary for hedgehog (Hh) pathway activation in MB cells. PDLIM3 is present in primary cilia of MB cells and fibroblasts, and such cilia localization is mediated by the PDZ domain of PDLIM3 protein. Deletion of PDLIM3 significantly compromised cilia formation and interfered the Hh signaling transduction in MB cells, suggesting that PDLIM3 promotes the Hh signaling through supporting the ciliogenesis. PDLIM3 protein physically interacts with cholesterol, a critical molecule for cilia formation and hedgehog signaling. The disruption of cilia formation and Hh signaling in PDLIM3 null MB cells or fibroblasts, was significantly rescued by treatment with exogenous cholesterol, demonstrating that PDLIM3 facilitates the ciliogenesis through cholesterol provision. Finally, deletion of PDLIM3 in MB cells significantly inhibited their proliferation and repressed tumor growth, suggesting that PDLIM3 is necessary for MB tumorigenesis. Our studies elucidate the critical functions of PDLIM3 in the ciliogenesis and Hh signaling transduction in SHH-MB cells, supporting to utilize PDLIM3 as a molecular marker for defining SHH group of MB in clinics.

Advances in Mixer Design and Detection Methods for Kinetics Studies of Macromolecular Folding and Binding on the Microsecond Time Scale.

Many important biological processes such as protein folding and ligand binding are too fast to be fully resolved using conventional stopped-flow techniques. Although advances in mixer design and detection methods have provided access to the microsecond time regime, there is room for improvement in terms of temporal resolution and sensitivity. To address this need, we developed a continuous-flow mixing instrument with a dead time of 12 to 27 µs (depending on solution viscosity) and enhanced sensitivity, sufficient for monitoring tryptophan or tyrosine fluorescence changes at fluorophore concentrations as low as 1 µM. Relying on commercially available laser microfabrication services, we obtained an integrated mixer/flow-cell assembly on a quartz chip, based on a cross-channel configuration with channel dimensions and geometry designed to minimize backpressure. By gradually increasing the width of the observation channel downstream from the mixing region, we are able to monitor a reaction progress time window ranging from ~10 µs out to ~3 ms. By combining a solid-state UV laser with a Galvano-mirror scanning strategy, we achieved highly efficient and uniform fluorescence excitation along the flow channel. Examples of applications, including refolding of acid-denatured cytochrome c triggered by a pH jump and binding of a peptide ligand to a PDZ domain, demonstrate the capability of the technique to resolve fluorescence changes down to the 10 µs time regime on modest amounts of reagents.

Semi-Quantitative MALDI Measurements of Blood-Based Samples for Molecular Diagnostics.

Accurate and precise measurement of the relative protein content of blood-based samples using mass spectrometry is challenging due to the large number of circulating proteins and the dynamic range of their abundances. Traditional spectral processing methods often struggle with accurately detecting overlapping peaks that are observed in these samples. In this work, we develop a novel spectral processing algorithm that effectively detects over 1650 peaks with over 3.5 orders of magnitude in intensity in the 3 to 30 kD m/z range. The algorithm utilizes a convolution of the peak shape to enhance peak detection, and accurate peak fitting to provide highly reproducible relative abundance estimates for both isolated peaks and overlapping peaks. We demonstrate a substantial increase in the reproducibility of the measurements of relative protein abundance when comparing this processing method to a traditional processing method for sample sets run on multiple matrix-assisted laser desorption/ionization-time of flight (MALDI-TOF) instruments. By utilizing protein set enrichment analysis, we find a sizable increase in the number of features associated with biological processes compared to previously reported results. The new processing method could be very beneficial when developing high-performance molecular diagnostic tests in disease indications.

Effects of ionic strength on the folding and stability of SAMP1, a ubiquitin-like halophilic protein.

Our knowledge of the folding behavior of proteins from extremophiles is limited at this time. These proteins may more closely resemble the primordial proteins selected in early evolution under extreme conditions. The small archaeal modifier protein 1 (SAMP1) studied in this report is an 87-residue protein with a ß-grasp fold found in the halophile Haloferax volcanii from the Dead Sea. To gain insight into the effects of salt on the stability and folding mechanism of SAMP1, we conducted equilibrium and kinetic folding experiments as a function of sodium chloride concentration. The results revealed that increasing ionic strength accelerates refolding and slows down unfolding of SAMP1, giving rise to a pronounced salt-induced stabilization. With increasing NaCl concentration, the rate of folding observed via a combination of continuous-flow (0.1-2 ms time range) and stopped-flow measurements (>2 ms) exhibited a >100-fold increase between 0.1 and 1.5 M NaCl and leveled off at higher concentrations. Using the Linderström-Lang smeared charge formalism to model electrostatic interactions in ground and transition states encountered during folding, we showed that the observed salt dependence is dominated by Debye-Hückel screening of electrostatic repulsion among numerous negatively charged residues. Comparisons are also drawn with three well-studied mesophilic members of the ß-grasp superfamily: protein G, protein L, and ubiquitin. Interestingly, the folding rate of SAMP1 in 3 M sodium chloride is comparable to that of protein G, ubiquitin, and protein L at lower ionic strength. The results indicate the important role of electrostatic interactions in protein folding and imply that proteins have evolved to minimize unfavorable charge-charge interactions under their specific native conditions.

Predicting prognosis in COVID-19 patients using machine learning and readily available clinical data.

RATIONALE: Prognostic tools for aiding in the treatment of hospitalized COVID-19 patients could help improve outcome by identifying patients at higher or lower risk of severe disease. The study objective was to develop models to stratify patients by risk of severe outcomes during COVID-19 hospitalization using readily available information at hospital admission. METHODS: Hierarchical ensemble classification models were trained on a set of 229 patients hospitalized with COVID-19 to predict severe outcomes, including ICU admission, development of acute respiratory distress syndrome, or intubation, using easily attainable attributes including basic patient characteristics, vital signs at admission, and basic lab results collected at time of presentation. Each test stratifies patients into groups of increasing risk. An additional cohort of 330 patients was used for blinded, independent validation. Shapley value analysis evaluated which attributes contributed most to the models' predictions of risk. MAIN RESULTS: Test performance was assessed using precision (positive predictive value) and recall (sensitivity) of the final risk groups. All test cut-offs were fixed prior to blinded validation. In development and validation, the tests achieved precision in the lowest risk groups near or above 0.9. The proportion of patients with severe outcomes significantly increased across increasing risk groups. While the importance of attributes varied by test and patient, C-reactive protein, lactate dehydrogenase, and D-dimer were often found to be important in the assignment of risk. CONCLUSIONS: Risk of severe outcomes for patients hospitalized with COVID-19 infection can be assessed using machine learning-based models based on attributes routinely collected at hospital admission.

Detection of Hepatocellular Carcinoma in a High-Risk Population by a Mass Spectrometry-Based Test.

Hepatocellular carcinoma (HCC) is one of the fastest growing causes of cancer-related death. Guidelines recommend obtaining a screening ultrasound with or without alpha-fetoprotein (AFP) every 6 months in at-risk adults. AFP as a screening biomarker is plagued by low sensitivity/specificity, prompting interest in discovering alternatives. Mass spectrometry-based techniques are promising in their ability to identify potential biomarkers. This study aimed to use machine learning utilizing spectral data and AFP to create a model for early detection. Serum samples were collected from three separate cohorts, and data were compiled to make Development, Internal Validation, and Independent Validation sets. AFP levels were measured, and Deep MALDI® analysis was used to generate mass spectra. Spectral data were input into the VeriStrat® classification algorithm. Machine learning techniques then classified each sample as "Cancer" or "No Cancer". Sensitivity and specificity of the test were >80% to detect HCC. High specificity of the test was independent of cause and severity of underlying disease. When compared to AFP, there was improved cancer detection for all tumor sizes, especially small lesions. Overall, a machine learning algorithm incorporating mass spectral data and AFP values from serum samples offers a novel approach to diagnose HCC. Given the small sample size of the Independent Validation set, a further independent, prospective study is warranted.

Explaining multivariate molecular diagnostic tests via Shapley values.

BACKGROUND: Machine learning (ML) can be an effective tool to extract information from attribute-rich molecular datasets for the generation of molecular diagnostic tests. However, the way in which the resulting scores or classifications are produced from the input data may not be transparent. Algorithmic explainability or interpretability has become a focus of ML research. Shapley values, first introduced in game theory, can provide explanations of the result generated from a specific set of input data by a complex ML algorithm. METHODS: For a multivariate molecular diagnostic test in clinical use (the VeriStrat® test), we calculate and discuss the interpretation of exact Shapley values. We also employ some standard approximation techniques for Shapley value computation (local interpretable model-agnostic explanation (LIME) and Shapley Additive Explanations (SHAP) based methods) and compare the results with exact Shapley values. RESULTS: Exact Shapley values calculated for data collected from a cohort of 256 patients showed that the relative importance of attributes for test classification varied by sample. While all eight features used in the VeriStrat® test contributed equally to classification for some samples, other samples showed more complex patterns of attribute importance for classification generation. Exact Shapley values and Shapley-based interaction metrics were able to provide interpretable classification explanations at the sample or patient level, while patient subgroups could be defined by comparing Shapley value profiles between patients. LIME and SHAP approximation approaches, even those seeking to include correlations between attributes, produced results that were quantitatively and, in some cases qualitatively, different from the exact Shapley values. CONCLUSIONS: Shapley values can be used to determine the relative importance of input attributes to the result generated by a multivariate molecular diagnostic test for an individual sample or patient. Patient subgroups defined by Shapley value profiles may motivate translational research. However, correlations inherent in molecular data and the typically small ML training sets available for molecular diagnostic test development may cause some approximation methods to produce approximate Shapley values that differ both qualitatively and quantitatively from exact Shapley values. Hence, caution is advised when using approximate methods to evaluate Shapley explanations of the results of molecular diagnostic tests.

A Phase 2 Randomized Placebo-Controlled Adjuvant Trial of GI-4000, a Recombinant Yeast Expressing Mutated RAS Proteins in Patients with Resected Pancreas Cancer.

Purpose: GI-4000, a series of recombinant yeast expressing four different mutated RAS proteins, was evaluated in subjects with resected ras-mutated pancreas cancer. Methods: Subjects (n = 176) received GI-4000 or placebo plus gemcitabine. Subjects' tumors were genotyped to identify which matched GI-4000 product to administer. Immune responses were measured by interferon-γ (IFNγ) ELISpot assay and by regulatory T cell (Treg) frequencies on treatment. Pretreatment plasma was retrospectively analyzed by matrix-assisted laser desorption/ionization-time-of-flight (MALDI-ToF) mass spectrometry for proteomic signatures predictive of GI-4000 responsiveness. Results: GI-4000 was well tolerated, with comparable safety findings between treatment groups. The GI-4000 group showed a similar pattern of median recurrence-free and overall survival (OS) compared with placebo. For the prospectively defined and stratified R1 resection subgroup, there was a trend in 1 year OS (72% vs. 56%), an improvement in OS (523.5 vs. 443.5 days [hazard ratio (HR) = 1.06 [confidence interval (CI): 0.53-2.13], p = 0.872), and increased frequency of immune responders (40% vs. 8%; p = 0.062) for GI-4000 versus placebo and a 159-day improvement in OS for R1 GI-4000 immune responders versus placebo (p = 0.810). For R0 resection subjects, no increases in IFNγ responses in GI-4000-treated subjects were observed. A higher frequency of R0/R1 subjects with a reduction in Tregs (CD4+/CD45RA+/Foxp3low) was observed in GI-4000-treated subjects versus placebo (p = 0.033). A proteomic signature was identified that predicted response to GI-4000/gemcitabine regardless of resection status. Conclusion: These results justify continued investigation of GI-4000 in studies stratified for likely responders or in combination with immune check-point inhibitors or other immunomodulators, which may provide optimal reactivation of antitumor immunity. ClinicalTrials.gov Number: NCT00300950.

Definition and Independent Validation of a Proteomic-Classifier in Ovarian Cancer.

Mass-spectrometry-based analyses have identified a variety of candidate protein biomarkers that might be crucial for epithelial ovarian cancer (EOC) development and therapy response. Comprehensive validation studies of the biological and clinical implications of proteomics are needed to advance them toward clinical use. Using the Deep MALDI method of mass spectrometry, we developed and independently validated (development cohort: n = 199, validation cohort: n = 135) a blood-based proteomic classifier, stratifying EOC patients into good and poor survival groups. We also determined an age dependency of the prognostic performance of this classifier, and our protein set enrichment analysis showed that the good and poor proteomic phenotypes were associated with, respectively, lower and higher levels of complement activation, inflammatory response, and acute phase reactants. This work highlights that, just like molecular markers of the tumor itself, the systemic condition of a patient (partly reflected in proteomic patterns) also influences survival and therapy response in a subset of ovarian cancer patients and could therefore be integrated into future processes of therapy planning.

Energetics and kinetics of substrate analog-coupled staphylococcal nuclease folding revealed by a statistical mechanical approach.

Protein conformational changes associated with ligand binding, especially those involving intrinsically disordered proteins, are mediated by tightly coupled intra- and intermolecular events. Such reactions are often discussed in terms of two limiting kinetic mechanisms, conformational selection (CS), where folding precedes binding, and induced fit (IF), where binding precedes folding. It has been shown that coupled folding/binding reactions can proceed along both CS and IF pathways with the flux ratio depending on conditions such as ligand concentration. However, the structural and energetic basis of such complex reactions remains poorly understood. Therefore, we used experimental, theoretical, and computational approaches to explore structural and energetic aspects of the coupled-folding/binding reaction of staphylococcal nuclease in the presence of the substrate analog adenosine-3',5'-diphosphate. Optically monitored equilibrium and kinetic data, combined with a statistical mechanical model, gave deeper insight into the relative importance of specific and Coulombic protein-ligand interactions in governing the reaction mechanism. We also investigated structural aspects of the reaction at the residue level using NMR and all-atom replica-permutation molecular dynamics simulations. Both approaches yielded clear evidence for accumulation of a transient protein-ligand encounter complex early in the reaction under IF-dominant conditions. Quantitative analysis of the equilibrium/kinetic folding revealed that the ligand-dependent CS-to-IF shift resulted from stabilization of the compact transition state primarily by weakly ligand-dependent Coulombic interactions with smaller contributions from specific binding energies. At a more macroscopic level, the CS-to-IF shift was represented as a displacement of the reaction "route" on the free energy surface, which was consistent with a flux analysis.

A Serum Protein Classifier Identifying Patients with Advanced Non-Small Cell Lung Cancer Who Derive Clinical Benefit from Treatment with Immune Checkpoint Inhibitors.

PURPOSE: Pretreatment selection of patients with non-small cell lung cancer (NSCLC) who would derive clinical benefit from treatment with immune checkpoint inhibitors (CPIs) would fulfill an unmet clinical need by reducing unnecessary toxicities from treatment and result in substantial health care savings. EXPERIMENTAL DESIGN: In a retrospective study, mass spectrometry (MS)-based proteomic analysis was performed on pretreatment sera derived from patients with advanced NSCLC treated with nivolumab as part of routine clinical care (n = 289). Machine learning combined spectral and clinical data to stratify patients into three groups with good ("sensitive"), intermediate, and poor ("resistant") outcomes following treatment in the second-line setting. The test was applied to three independent patient cohorts and its biology was investigated using protein set enrichment analyses (PSEA). RESULTS: A signature consisting of 274 MS features derived from a development set of 116 patients was associated with progression-free survival (PFS) and overall survival (OS) across two validation cohorts (N = 98 and N = 75). In pooled analysis, significantly better OS was demonstrated for "sensitive" relative to "not sensitive" patients treated with nivolumab; HR, 0.58 (95% confidence interval, 0.38-0-87; P = 0.009). There was no significant association with clinical factors including PD-L1 expression, available from 133 of 289 patients. The test demonstrated no significant association with PFS or OS in a historical cohort (n = 68) of second-line NSCLC patients treated with docetaxel. PSEA revealed proteomic classification to be significantly associated with complement and wound-healing cascades. CONCLUSIONS: This serum-derived protein signature successfully stratified outcomes in cohorts of patients with advanced NSCLC treated with second-line PD-1 CPIs and deserves further prospective study.

Gene network transitions in embryos depend upon interactions between a pioneer transcription factor and core histones.

Gene network transitions in embryos and other fate-changing contexts involve combinations of transcription factors. A subset of fate-changing transcription factors act as pioneers; they scan and target nucleosomal DNA and initiate cooperative events that can open the local chromatin. However, a gap has remained in understanding how molecular interactions with the nucleosome contribute to the chromatin-opening phenomenon. Here we identified a short α-helical region, conserved among FOXA pioneer factors, that interacts with core histones and contributes to chromatin opening in vitro. The same domain is involved in chromatin opening in early mouse embryos for normal development. Thus, local opening of chromatin by interactions between pioneer factors and core histones promotes genetic programming.

Mass Spectrometry-Based Multivariate Proteomic Tests for Prediction of Outcomes on Immune Checkpoint Blockade Therapy: The Modern Analytical Approach.

The remarkable success of immune checkpoint inhibitors (ICIs) has given hope of cure for some patients with advanced cancer; however, the fraction of responding patients is 15-35%, depending on tumor type, and the proportion of durable responses is even smaller. Identification of biomarkers with strong predictive potential remains a priority. Until now most of the efforts were focused on biomarkers associated with the assumed mechanism of action of ICIs, such as levels of expression of programmed death-ligand 1 (PD-L1) and mutation load in tumor tissue, as a proxy of immunogenicity; however, their performance is unsatisfactory. Several assays designed to capture the complexity of the disease by measuring the immune response in tumor microenvironment show promise but still need validation in independent studies. The circulating proteome contains an additional layer of information characterizing tumor-host interactions that can be integrated into multivariate tests using modern machine learning techniques. Here we describe several validated serum-based proteomic tests and their utility in the context of ICIs. We discuss test performances, demonstrate their independence from currently used biomarkers, and discuss various aspects of associated biological mechanisms. We propose that serum-based multivariate proteomic tests add a missing piece to the puzzle of predicting benefit from ICIs.

A proposal for score assignment to characterize biological processes from mass spectral analysis of serum.

INTRODUCTION: Most diseases involve a complex interplay between multiple biological processes at the cellular, tissue, organ, and systemic levels. Clinical tests and biomarkers based on the measurement of a single or few analytes may not be able to capture the complexity of a patient's disease. Novel approaches for comprehensively assessing biological processes from easily obtained samples could help in the monitoring, treatment, and understanding of many conditions. OBJECTIVES: We propose a method of creating scores associated with specific biological processes from mass spectral analysis of serum samples. METHODS: A score for a process of interest is created by: (i) identifying mass spectral features associated with the process using set enrichment analysis methods, and (ii) combining these features into a score using a principal component analysis-based approach. We investigate the creation of scores using cohorts of patients with non-small cell lung cancer, melanoma, and ovarian cancer. Since the circulating proteome is amenable to the study of immune responses, which play a critical role in cancer development and progression, we focus on functions related to the host response to disease. RESULTS: We demonstrate the feasibility of generating scores, their reproducibility, and their associations with clinical outcomes. Once the scores are constructed, only 3 µL of serum is required for the assessment of multiple biological functions from the circulating proteome. CONCLUSION: These mass spectrometry-based scores could be useful for future multivariate biomarker or test development studies for informing treatment, disease monitoring and improving understanding of the roles of various biological functions in multiple disease settings.

Extending the information content of the MALDI analysis of biological fluids via multi-million shot analysis.

INTRODUCTION: Reliable measurements of the protein content of biological fluids like serum or plasma can provide valuable input for the development of personalized medicine tests. Standard MALDI analysis typically only shows high abundance proteins, which limits its utility for test development. It also exhibits reproducibility issues with respect to quantitative measurements. In this paper we show how the sensitivity of MALDI profiling of intact proteins in unfractionated human serum can be substantially increased by exposing a sample to many more laser shots than are commonly used. Analytical reproducibility is also improved. METHODS: To assess what is theoretically achievable we utilized spectra from the same samples obtained over many years and combined them to generate MALDI spectral averages of up to 100,000,000 shots for a single sample, and up to 8,000,000 shots for a set of 40 different serum samples. Spectral attributes, such as number of peaks and spectral noise of such averaged spectra were investigated together with analytical reproducibility as a function of the number of shots. We confirmed that results were similar on MALDI instruments from different manufacturers. RESULTS: We observed an expected decrease of noise, roughly proportional to the square root of the number of shots, over the whole investigated range of the number of shots (5 orders of magnitude), resulting in an increase in the number of reliably detected peaks. The reproducibility of the amplitude of these peaks, measured by CV and concordance analysis also improves with very similar dependence on shot number, reaching median CVs below 2% for shot numbers > 4 million. Measures of analytical information content and association with biological processes increase with increasing number of shots. CONCLUSIONS: We demonstrate that substantially increasing the number of laser shots in a MALDI-TOF analysis leads to more informative and reliable data on the protein content of unfractionated serum. This approach has already been used in the development of clinical tests in oncology.

A dropout-regularized classifier development approach optimized for precision medicine test discovery from omics data.

BACKGROUND: Modern genomic and proteomic profiling methods produce large amounts of data from tissue and blood-based samples that are of potential utility for improving patient care. However, the design of precision medicine tests for unmet clinical needs from this information in the small cohorts available for test discovery remains a challenging task. Obtaining reliable performance assessments at the earliest stages of test development can also be problematic. We describe a novel approach to classifier development designed to create clinically useful tests together with reliable estimates of their performance. The method incorporates elements of traditional and modern machine learning to facilitate the use of cohorts where the number of samples is less than the number of measured patient attributes. It is based on a hierarchy of classification and information abstraction and combines boosting, bagging, and strong dropout regularization. RESULTS: We apply this dropout-regularized combination approach to two clinical problems in oncology using mRNA expression and associated clinical data and compare performance with other methods of classifier generation, including Random Forest. Performance of the new method is similar to or better than the Random Forest in the two classification tasks used for comparison. The dropout-regularized combination method also generates an effective classifier in a classification task with a known confounding variable. Most importantly, it provides a reliable estimate of test performance from a relatively small development set of samples. CONCLUSIONS: The flexible dropout-regularized combination approach is able to produce tests tailored to particular clinical questions and mitigate known confounding effects. It allows the design of molecular diagnostic tests addressing particular clinical questions together with reliable assessment of whether test performance is likely to be fit-for-purpose in independent validation at the earliest stages of development.

Robust identification of molecular phenotypes using semi-supervised learning.

BACKGROUND: Modern molecular profiling techniques are yielding vast amounts of data from patient samples that could be utilized with machine learning methods to provide important biological insights and improvements in patient outcomes. Unsupervised methods have been successfully used to identify molecularly-defined disease subtypes. However, these approaches do not take advantage of potential additional clinical outcome information. Supervised methods can be implemented when training classes are apparent (e.g., responders or non-responders to treatment). However, training classes can be difficult to define when assessing relative benefit of one therapy over another using gold standard clinical endpoints, since it is often not clear how much benefit each individual patient receives. RESULTS: We introduce an iterative approach to binary classification tasks based on the simultaneous refinement of training class labels and classifiers towards self-consistency. As training labels are refined during the process, the method is well suited to cases where training class definitions are not obvious or noisy. Clinical data, including time-to-event endpoints, can be incorporated into the approach to enable the iterative refinement to identify molecular phenotypes associated with a particular clinical variable. Using synthetic data, we show how this approach can be used to increase the accuracy of identification of outcome-related phenotypes and their associated molecular attributes. Further, we demonstrate that the advantages of the method persist in real world genomic datasets, allowing the reliable identification of molecular phenotypes and estimation of their association with outcome that generalizes to validation datasets. We show that at convergence of the iterative refinement, there is a consistent incorporation of the molecular data into the classifier yielding the molecular phenotype and that this allows a robust identification of associated attributes and the underlying biological processes. CONCLUSIONS: The consistent incorporation of the structure of the molecular data into the classifier helps to minimize overfitting and facilitates not only good generalization of classification and molecular phenotypes, but also reliable identification of biologically relevant features and elucidation of underlying biological processes.

Proteomic test for anti-PD-1 checkpoint blockade treatment of metastatic melanoma with and without BRAF mutations.

The therapeutic landscape in metastatic melanoma has changed dramatically in the last decade, with the success of immune checkpoint inhibitors resulting in durable responses for a large number of patients. For patients with BRAF mutations, combinations of BRAF and MEK inhibitors demonstrated response rates and benefit comparable to those from immune checkpoint inhibitors, providing the rationale for sequential treatment with targeted and immunotherapies and raising the question of optimal treatment sequencing.Biomarkers for the selection of anti-PD-1 therapy in BRAF wild type (BRAF WT) and in BRAF mutated (BRAF MUT) patients help development of alternative treatments for patients unlikely to benefit, and might lead to better understanding of the interaction of checkpoint inhibition and targeted therapy. In this paper we evaluate the performance of a previously developed serum proteomic test, BDX008, in metastatic melanoma patients treated with anti-PD-1 agents and investigate the role of BRAF mutation status. BDX008, a pre-treatment proteomic test associated with acute phase reactants, wound healing and complement activation, stratifies patients into two groups, BDX008+ and BDX008-, with better and worse outcomes on immunotherapy.Serum samples were available from 71 patients treated with anti-PD1 inhibitors; 25 patients had BRAF mutations, 39 were wild type. Overall, BDX008+ patients had significantly better overall survival (OS) (HR = 0.50, P = 0.016) and a trend for better progression-free survival (PFS) (HR = 0.61, P = 0.060) than BDX008- patients. BDX008 classification was statistically significant in the analyses adjusted for mutation status, LDH, and line of treatment (P = 0.009 for OS and 0.031 for PFS). BRAF WT BDX008+ patients had markedly long median OS of 32.5 months and 53% landmark 2 years survival, with statistically significantly superior OS as compared to BDX008- patients (HR = 0.41, P = 0.032). The difference between BDX008+ and BDX008- in PFS in BRAF WT patients and in OS and PFS in BRAF MUT patients did not reach statistical significance, though numerically was consistent with overall results. The test demonstrated significant interaction with neutrophil-to-lymphocyte ratio (NLR) (PFS P = 0.041, OS P = 0.004). BDX008 as a biomarker selecting for benefit from immune checkpoint blockade, especially in patients with wild type BRAF and in subgroups with low NLR, warrants further evaluation.