Genetic program activity delineates risk, relapse, and therapy responsiveness in multiple myeloma.

Despite recent advancements in the treatment of multiple myeloma (MM), nearly all patients ultimately relapse and many become refractory to multiple lines of therapies. Therefore, we not only need the ability to predict which patients are at high risk for disease progression but also a means to understand the mechanisms underlying their risk. Here, we report a transcriptional regulatory network (TRN) for MM inferred from cross-sectional multi-omics data from 881 patients that predicts how 124 chromosomal abnormalities and somatic mutations causally perturb 392 transcription regulators of 8549 genes to manifest in distinct clinical phenotypes and outcomes. We identified 141 genetic programs whose activity profiles stratify patients into 25 distinct transcriptional states and proved to be more predictive of outcomes than did mutations. The coherence of these programs and accuracy of our network-based risk prediction was validated in two independent datasets. We observed subtype-specific vulnerabilities to interventions with existing drugs and revealed plausible mechanisms for relapse, including the establishment of an immunosuppressive microenvironment. Investigation of the t(4;14) clinical subtype using the TRN revealed that 16% of these patients exhibit an extreme-risk combination of genetic programs (median progression-free survival of 5 months) that create a distinct phenotype with targetable genes and pathways.

A systems-level gene regulatory network model for Plasmodium falciparum.

Many of the gene regulatory processes of Plasmodium falciparum, the deadliest malaria parasite, remain poorly understood. To develop a comprehensive guide for exploring this organism's gene regulatory network, we generated a systems-level model of P. falciparum gene regulation using a well-validated, machine-learning approach for predicting interactions between transcription regulators and their targets. The resulting network accurately predicts expression levels of transcriptionally coherent gene regulatory programs in independent transcriptomic data sets from parasites collected by different research groups in diverse laboratory and field settings. Thus, our results indicate that our gene regulatory model has predictive power and utility as a hypothesis-generating tool for illuminating clinically relevant gene regulatory mechanisms within P. falciparum. Using the set of regulatory programs we identified, we also investigated correlates of artemisinin resistance based on gene expression coherence. We report that resistance is associated with incoherent expression across many regulatory programs, including those controlling genes associated with erythrocyte-host engagement. These results suggest that parasite populations with reduced artemisinin sensitivity are more transcriptionally heterogenous. This pattern is consistent with a model where the parasite utilizes bet-hedging strategies to diversify the population, rendering a subpopulation more able to navigate drug treatment.

Bone marrow microenvironments that contribute to patient outcomes in newly diagnosed multiple myeloma: A cohort study of patients in the Total Therapy clinical trials.

BACKGROUND: The tumor microenvironment (TME) is increasingly appreciated as an important determinant of cancer outcome, including in multiple myeloma (MM). However, most myeloma microenvironment studies have been based on bone marrow (BM) aspirates, which often do not fully reflect the cellular content of BM tissue itself. To address this limitation in myeloma research, we systematically characterized the whole bone marrow (WBM) microenvironment during premalignant, baseline, on treatment, and post-treatment phases. METHODS AND FINDINGS: Between 2004 and 2019, 998 BM samples were taken from 436 patients with newly diagnosed MM (NDMM) at the University of Arkansas for Medical Sciences in Little Rock, Arkansas, United States of America. These patients were 61% male and 39% female, 89% White, 8% Black, and 3% other/refused, with a mean age of 58 years. Using WBM and matched cluster of differentiation (CD)138-selected tumor gene expression to control for tumor burden, we identified a subgroup of patients with an adverse TME associated with 17 fewer months of progression-free survival (PFS) (95% confidence interval [CI] 5-29, 49-69 versus 70-82 months, χ2 p = 0.001) and 15 fewer months of overall survival (OS; 95% CI -1 to 31, 92-120 versus 113-129 months, χ2 p = 0.036). Using immunohistochemistry-validated computational tools that identify distinct cell types from bulk gene expression, we showed that the adverse outcome was correlated with elevated CD8+ T cell and reduced granulocytic cell proportions. This microenvironment develops during the progression of premalignant to malignant disease and becomes less prevalent after therapy, in which it is associated with improved outcomes. In patients with quantified International Staging System (ISS) stage and 70-gene Prognostic Risk Score (GEP-70) scores, taking the microenvironment into consideration would have identified an additional 40 out of 290 patients (14%, premutation p = 0.001) with significantly worse outcomes (PFS, 95% CI 6-36, 49-73 versus 74-90 months) who were not identified by existing clinical (ISS stage III) and tumor (GEP-70) criteria as high risk. The main limitations of this study are that it relies on computationally identified cell types and that patients were treated with thalidomide rather than current therapies. CONCLUSIONS: In this study, we observe that granulocyte signatures in the MM TME contribute to a more accurate prognosis. This implies that future researchers and clinicians treating patients should quantify TME components, in particular monocytes and granulocytes, which are often ignored in microenvironment studies.

A generalizable data-driven multicellular model of pancreatic ductal adenocarcinoma.

BACKGROUND: Mechanistic models, when combined with pertinent data, can improve our knowledge regarding important molecular and cellular mechanisms found in cancer. These models make the prediction of tissue-level response to drug treatment possible, which can lead to new therapies and improved patient outcomes. Here we present a data-driven multiscale modeling framework to study molecular interactions between cancer, stromal, and immune cells found in the tumor microenvironment. We also develop methods to use molecular data available in The Cancer Genome Atlas to generate sample-specific models of cancer. RESULTS: By combining published models of different cells relevant to pancreatic ductal adenocarcinoma (PDAC), we built an agent-based model of the multicellular pancreatic tumor microenvironment, formally describing cell type-specific molecular interactions and cytokine-mediated cell-cell communications. We used an ensemble-based modeling approach to systematically explore how variations in the tumor microenvironment affect the viability of cancer cells. The results suggest that the autocrine loop involving EGF signaling is a key interaction modulator between pancreatic cancer and stellate cells. EGF is also found to be associated with previously described subtypes of PDAC. Moreover, the model allows a systematic exploration of the effect of possible therapeutic perturbations; our simulations suggest that reducing bFGF secretion by stellate cells will have, on average, a positive impact on cancer apoptosis. CONCLUSIONS: The developed framework allows model-driven hypotheses to be generated regarding therapeutically relevant PDAC states with potential molecular and cellular drivers indicating specific intervention strategies.

Multiple Myeloma DREAM Challenge reveals epigenetic regulator PHF19 as marker of aggressive disease.

While the past decade has seen meaningful improvements in clinical outcomes for multiple myeloma patients, a subset of patients does not benefit from current therapeutics for unclear reasons. Many gene expression-based models of risk have been developed, but each model uses a different combination of genes and often involves assaying many genes making them difficult to implement. We organized the Multiple Myeloma DREAM Challenge, a crowdsourced effort to develop models of rapid progression in newly diagnosed myeloma patients and to benchmark these against previously published models. This effort lead to more robust predictors and found that incorporating specific demographic and clinical features improved gene expression-based models of high risk. Furthermore, post-challenge analysis identified a novel expression-based risk marker, PHF19, which has recently been found to have an important biological role in multiple myeloma. Lastly, we show that a simple four feature predictor composed of age, ISS, and expression of PHF19 and MMSET performs similarly to more complex models with many more gene expression features included.

ADAPTS: Automated deconvolution augmentation of profiles for tissue specific cells.

Immune cell infiltration of tumors and the tumor microenvironment can be an important component for determining patient outcomes. For example, immune and stromal cell presence inferred by deconvolving patient gene expression data may help identify high risk patients or suggest a course of treatment. One particularly powerful family of deconvolution techniques uses signature matrices of genes that uniquely identify each cell type as determined from single cell type purified gene expression data. Many methods from this family have been recently published, often including new signature matrices appropriate for a single purpose, such as investigating a specific type of tumor. The package ADAPTS helps users make the most of this expanding knowledge base by introducing a framework for cell type deconvolution. ADAPTS implements modular tools for customizing signature matrices for new tissue types by adding custom cell types or building new matrices de novo, including from single cell RNAseq data. It includes a common interface to several popular deconvolution algorithms that use a signature matrix to estimate the proportion of cell types present in heterogenous samples. ADAPTS also implements a novel method for clustering cell types into groups that are difficult to distinguish by deconvolution and then re-splitting those clusters using hierarchical deconvolution. We demonstrate that the techniques implemented in ADAPTS improve the ability to reconstruct the cell types present in a single cell RNAseq data set in a blind predictive analysis. ADAPTS is currently available for use in R on CRAN and GitHub.

The Promise of Systems Biology Approaches for Revealing Host Pathogen Interactions in Malaria.

Despite global eradication efforts over the past century, malaria remains a devastating public health burden, causing almost half a million deaths annually (WHO, 2016). A detailed understanding of the mechanisms that control malaria infection has been hindered by technical challenges of studying a complex parasite life cycle in multiple hosts. While many interventions targeting the parasite have been implemented, the complex biology of Plasmodium poses a major challenge, and must be addressed to enable eradication. New approaches for elucidating key host-parasite interactions, and predicting how the parasite will respond in a variety of biological settings, could dramatically enhance the efficacy and longevity of intervention strategies. The field of systems biology has developed methodologies and principles that are well poised to meet these challenges. In this review, we focus our attention on the Liver Stage of the Plasmodium lifecycle and issue a "call to arms" for using systems biology approaches to forge a new era in malaria research. These approaches will reveal insights into the complex interplay between host and pathogen, and could ultimately lead to novel intervention strategies that contribute to malaria eradication.

Linking EPCR-Binding PfEMP1 to Brain Swelling in Pediatric Cerebral Malaria.

Brain swelling is a major predictor of mortality in pediatric cerebral malaria (CM). However, the mechanisms leading to swelling remain poorly defined. Here, we combined neuroimaging, parasite transcript profiling, and laboratory blood profiles to develop machine-learning models of malarial retinopathy and brain swelling. We found that parasite var transcripts encoding endothelial protein C receptor (EPCR)-binding domains, in combination with high parasite biomass and low platelet levels, are strong indicators of CM cases with malarial retinopathy. Swelling cases presented low platelet levels and increased transcript abundance of parasite PfEMP1 DC8 and group A EPCR-binding domains. Remarkably, the dominant transcript in 50% of swelling cases encoded PfEMP1 group A CIDRα1.7 domains. Furthermore, a recombinant CIDRα1.7 domain from a pediatric CM brain autopsy inhibited the barrier-protective properties of EPCR in human brain endothelial cells in vitro. Together, these findings suggest a detrimental role for EPCR-binding CIDRα1 domains in brain swelling.

An indicator cell assay for blood-based diagnostics.

We have established proof of principle for the Indicator Cell Assay Platform™ (iCAP™), a broadly applicable tool for blood-based diagnostics that uses specifically-selected, standardized cells as biosensors, relying on their innate ability to integrate and respond to diverse signals present in patients' blood. To develop an assay, indicator cells are exposed in vitro to serum from case or control subjects and their global differential response patterns are used to train reliable, disease classifiers based on a small number of features. In a feasibility study, the iCAP detected pre-symptomatic disease in a murine model of amyotrophic lateral sclerosis (ALS) with 94% accuracy (p-Value = 3.81E-6) and correctly identified samples from a murine Huntington's disease model as non-carriers of ALS. Beyond the mouse model, in a preliminary human disease study, the iCAP detected early stage Alzheimer's disease with 72% cross-validated accuracy (p-Value = 3.10E-3). For both assays, iCAP features were enriched for disease-related genes, supporting the assay's relevance for disease research.

Combining inferred regulatory and reconstructed metabolic networks enhances phenotype prediction in yeast.

Gene regulatory and metabolic network models have been used successfully in many organisms, but inherent differences between them make networks difficult to integrate. Probabilistic Regulation Of Metabolism (PROM) provides a partial solution, but it does not incorporate network inference and underperforms in eukaryotes. We present an Integrated Deduced And Metabolism (IDREAM) method that combines statistically inferred Environment and Gene Regulatory Influence Network (EGRIN) models with the PROM framework to create enhanced metabolic-regulatory network models. We used IDREAM to predict phenotypes and genetic interactions between transcription factors and genes encoding metabolic activities in the eukaryote, Saccharomyces cerevisiae. IDREAM models contain many fewer interactions than PROM and yet produce significantly more accurate growth predictions. IDREAM consistently outperformed PROM using any of three popular yeast metabolic models and across three experimental growth conditions. Importantly, IDREAM's enhanced accuracy makes it possible to identify subtle synthetic growth defects. With experimental validation, these novel genetic interactions involving the pyruvate dehydrogenase complex suggested a new role for fatty acid-responsive factor Oaf1 in regulating acetyl-CoA production in glucose grown cells.

Severe adult malaria is associated with specific PfEMP1 adhesion types and high parasite biomass.

The interplay between cellular and molecular determinants that lead to severe malaria in adults is unexplored. Here, we analyzed parasite virulence factors in an infected adult population in India and investigated whether severe malaria isolates impair endothelial protein C receptor (EPCR), a protein involved in coagulation and endothelial barrier permeability. Severe malaria isolates overexpressed specific members of the Plasmodium falciparum var gene/PfEMP1 (P. falciparum erythrocyte membrane protein 1) family that bind EPCR, including DC8 var genes that have previously been linked to severe pediatric malaria. Machine learning analysis revealed that DC6- and DC8-encoding var transcripts in combination with high parasite biomass were the strongest indicators of patient hospitalization and disease severity. We found that DC8 CIDRα1 domains from severe malaria isolates had substantial differences in EPCR binding affinity and blockade activity for its ligand activated protein C. Additionally, even a low level of inhibition exhibited by domains from two cerebral malaria isolates was sufficient to interfere with activated protein C-barrier protective activities in human brain endothelial cells. Our findings demonstrate an interplay between parasite biomass and specific PfEMP1 adhesion types in the development of adult severe malaria, and indicate that low impairment of EPCR function may contribute to parasite virulence.

Bicluster Sampled Coherence Metric (BSCM) provides an accurate environmental context for phenotype predictions.

BACKGROUND: Biclustering is a popular method for identifying under which experimental conditions biological signatures are co-expressed. However, the general biclustering problem is NP-hard, offering room to focus algorithms on specific biological tasks. We hypothesize that conditional co-regulation of genes is a key factor in determining cell phenotype and that accurately segregating conditions in biclusters will improve such predictions. Thus, we developed a bicluster sampled coherence metric (BSCM) for determining which conditions and signals should be included in a bicluster. RESULTS: Our BSCM calculates condition and cluster size specific p-values, and we incorporated these into the popular integrated biclustering algorithm cMonkey. We demonstrate that incorporation of our new algorithm significantly improves bicluster co-regulation scores (p-value = 0.009) and GO annotation scores (p-value = 0.004). Additionally, we used a bicluster based signal to predict whether a given experimental condition will result in yeast peroxisome induction. Using the new algorithm, the classifier accuracy improves from 41.9% to 76.1% correct. CONCLUSIONS: We demonstrate that the proposed BSCM helps determine which signals ought to be co-clustered, resulting in more accurately assigned bicluster membership. Furthermore, we show that BSCM can be extended to more accurately detect under which experimental conditions the genes are co-clustered. Features derived from this more accurate analysis of conditional regulation results in a dramatic improvement in the ability to predict a cellular phenotype in yeast. The latest cMonkey is available for download at https://github.com/baliga-lab/cmonkey2. The experimental data and source code featured in this paper is available http://AitchisonLab.com/BSCM. BSCM has been incorporated in the official cMonkey release.

Systematic Analysis of Yeast Proteome Reveals Peptide Detectability Factors for Mass Spectrometry.

Here we used a data-independent acquisition (DIA) method, Precursor Acquisition Independent from Ion Count (PAcIFIC), to systematically profile the S. cerevisiae proteome. Direct PAcIFIC analysis of a yeast whole cell lysate (WCL) yielded 90% reproducibility between replicates and detected approximately 2000 proteins. When combined with sub-cellular fractionation, reproducibility was equally high and the number of detected yeast proteins approached 5000. As noted previously, this unbiased DIA approach identified so-called "orphan" peptides that could only be detected by tandem mass spectra because there was no detectable precursor ion. Using this unique dataset we examined features associated with peptide detectability and demonstrated that orphans were more likely to arise from low copy number proteins than proteins with median or high copy number. Finally, an investigation into why some orphans also arose from high copy number proteins found that, aside from protein copy number, there was a bias toward physiochemical factors associated with regions flanking the proteolytic cleavage sites of orphan peptides. This suggested that those orphan peptides originating from high abundance proteins were likely the result of inefficient protease release, which has implications for quantitative bottom-up proteomics.

Predicting antidisease immunity using proteome arrays and sera from children naturally exposed to malaria.

Malaria remains one of the most prevalent and lethal human infectious diseases worldwide. A comprehensive characterization of antibody responses to blood stage malaria is essential to support the development of future vaccines, sero-diagnostic tests, and sero-surveillance methods. We constructed a proteome array containing 4441 recombinant proteins expressed by the blood stages of the two most common human malaria parasites, P. falciparum (Pf) and P. vivax (Pv), and used this array to screen sera of Papua New Guinea children infected with Pf, Pv, or both (Pf/Pv) that were either symptomatic (febrile), or asymptomatic but had parasitemia detectable via microscopy or PCR. We hypothesized that asymptomatic children would develop antigen-specific antibody profiles associated with antidisease immunity, as compared with symptomatic children. The sera from these children recognized hundreds of the arrayed recombinant Pf and Pv proteins. In general, responses in asymptomatic children were highest in those with high parasitemia, suggesting that antibody levels are associated with parasite burden. In contrast, symptomatic children carried fewer antibodies than asymptomatic children with infections detectable by microscopy, particularly in Pv and Pf/Pv groups, suggesting that antibody production may be impaired during symptomatic infections. We used machine-learning algorithms to investigate the relationship between antibody responses and symptoms, and we identified antibody responses to sets of Plasmodium proteins that could predict clinical status of the donors. Several of these antibody responses were identified by multiple comparisons, including those against members of the serine enriched repeat antigen family and merozoite protein 4. Interestingly, both P. falciparum serine enriched repeat antigen-5 and merozoite protein 4 have been previously investigated for use in vaccines. This machine learning approach, never previously applied to proteome arrays, can be used to generate a list of potential seroprotective and/or diagnostic antigens candidates that can be further evaluated in longitudinal studies.

Molecular mechanisms of system responses to novel stimuli are predictable from public data.

Systems scale models provide the foundation for an effective iterative cycle between hypothesis generation, experiment and model refinement. Such models also enable predictions facilitating the understanding of biological complexity and the control of biological systems. Here, we demonstrate the reconstruction of a globally predictive gene regulatory model from public data: a model that can drive rational experiment design and reveal new regulatory mechanisms underlying responses to novel environments. Specifically, using ∼ 1500 publically available genome-wide transcriptome data sets from Saccharomyces cerevisiae, we have reconstructed an environment and gene regulatory influence network that accurately predicts regulatory mechanisms and gene expression changes on exposure of cells to completely novel environments. Focusing on transcriptional networks that induce peroxisomes biogenesis, the model-guided experiments allow us to expand a core regulatory network to include novel transcriptional influences and linkage across signaling and transcription. Thus, the approach and model provides a multi-scalar picture of gene dynamics and are powerful resources for exploiting extant data to rationally guide experimentation. The techniques outlined here are generally applicable to any biological system, which is especially important when experimental systems are challenging and samples are difficult and expensive to obtain-a common problem in laboratory animal and human studies.

The functional role of SUMO E3 ligase Mms21p in the maintenance of subtelomeric silencing in budding yeast.

In Saccharomyces cerevisiae, subtelomeric silencing is involved in the propagation of Silent Information Regulator (SIR) proteins toward euchromatin. Numerous mechanisms are involved in antagonizing the local spread of Sir-dependent silent chromatin into neighboring euchromatin. Here, we identified a novel role for sumoylation E3 ligase Mms21 in the maintenance of subtelomeric silencing. We found that disruption of E3 ligase activity of Mms21 results in the de-repression of subtelomeric silencing. Deletion of E3 ligase domain of Mms21 led to decreased binding of Sir2p, Sir3p and Sir4 at subtelomeric chromatins and increased H3K4 tri-methylation at telomere-distal euchromatin regions, correlating with increased gene expression in two subtelomeric reporter genes. In addition, a mms21Δsl mutant caused a severe growth defect in combination with htz1Δ deletion and showed an enhanced association of Htz1 with telomere proximal regions. Taken together, our findings suggest an important role of Mms21p; it contributes to subtelomeric silencing during the formation of a heterochromatin boundary.

PhosphoChain: a novel algorithm to predict kinase and phosphatase networks from high-throughput expression data.

MOTIVATION: Protein phosphorylation is critical for regulating cellular activities by controlling protein activities, localization and turnover, and by transmitting information within cells through signaling networks. However, predictions of protein phosphorylation and signaling networks remain a significant challenge, lagging behind predictions of transcriptional regulatory networks into which they often feed. RESULTS: We developed PhosphoChain to predict kinases, phosphatases and chains of phosphorylation events in signaling networks by combining mRNA expression levels of regulators and targets with a motif detection algorithm and optional prior information. PhosphoChain correctly reconstructed ∼78% of the yeast mitogen-activated protein kinase pathway from publicly available data. When tested on yeast phosphoproteomic data from large-scale mass spectrometry experiments, PhosphoChain correctly identified ∼27% more phosphorylation sites than existing motif detection tools (NetPhosYeast and GPS2.0), and predictions of kinase-phosphatase interactions overlapped with ∼59% of known interactions present in yeast databases. PhosphoChain provides a valuable framework for predicting condition-specific phosphorylation events from high-throughput data. AVAILABILITY: PhosphoChain is implemented in Java and available at http://virgo.csie.ncku.edu.tw/PhosphoChain/ or http://aitchisonlab.com/PhosphoChain

All-codon scanning identifies p53 cancer rescue mutations.

In vitro scanning mutagenesis strategies are valuable tools to identify critical residues in proteins and to generate proteins with modified properties. We describe the fast and simple All-Codon Scanning (ACS) strategy that creates a defined gene library wherein each individual codon within a specific target region is changed into all possible codons with only a single codon change per mutagenesis product. ACS is based on a multiplexed overlapping mutagenesis primer design that saturates only the targeted gene region with single codon changes. We have used ACS to produce single amino-acid changes in small and large regions of the human tumor suppressor protein p53 to identify single amino-acid substitutions that can restore activity to inactive p53 found in human cancers. Single-tube reactions were used to saturate defined 30-nt regions with all possible codon changes. The same technique was used in 20 parallel reactions to scan the 600-bp fragment encoding the entire p53 core domain. Identification of several novel p53 cancer rescue mutations demonstrated the utility of the ACS approach. ACS is a fast, simple and versatile method, which is useful for protein structure-function analyses and protein design or evolution problems.

Predicting positive p53 cancer rescue regions using Most Informative Positive (MIP) active learning.

Many protein engineering problems involve finding mutations that produce proteins with a particular function. Computational active learning is an attractive approach to discover desired biological activities. Traditional active learning techniques have been optimized to iteratively improve classifier accuracy, not to quickly discover biologically significant results. We report here a novel active learning technique, Most Informative Positive (MIP), which is tailored to biological problems because it seeks novel and informative positive results. MIP active learning differs from traditional active learning methods in two ways: (1) it preferentially seeks Positive (functionally active) examples; and (2) it may be effectively extended to select gene regions suitable for high throughput combinatorial mutagenesis. We applied MIP to discover mutations in the tumor suppressor protein p53 that reactivate mutated p53 found in human cancers. This is an important biomedical goal because p53 mutants have been implicated in half of all human cancers, and restoring active p53 in tumors leads to tumor regression. MIP found Positive (cancer rescue) p53 mutants in silico using 33% fewer experiments than traditional non-MIP active learning, with only a minor decrease in classifier accuracy. Applying MIP to in vivo experimentation yielded immediate Positive results. Ten different p53 mutations found in human cancers were paired in silico with all possible single amino acid rescue mutations, from which MIP was used to select a Positive Region predicted to be enriched for p53 cancer rescue mutants. In vivo assays showed that the predicted Positive Region: (1) had significantly more (p<0.01) new strong cancer rescue mutants than control regions (Negative, and non-MIP active learning); (2) had slightly more new strong cancer rescue mutants than an Expert region selected for purely biological considerations; and (3) rescued for the first time the previously unrescuable p53 cancer mutant P152L.

Choosing where to look next in a mutation sequence space: Active Learning of informative p53 cancer rescue mutants.

MOTIVATION: Many biomedical projects would benefit from reducing the time and expense of in vitro experimentation by using computer models for in silico predictions. These models may help determine which expensive biological data are most useful to acquire next. Active Learning techniques for choosing the most informative data enable biologists and computer scientists to optimize experimental data choices for rapid discovery of biological function. To explore design choices that affect this desirable behavior, five novel and five existing Active Learning techniques, together with three control methods, were tested on 57 previously unknown p53 cancer rescue mutants for their ability to build classifiers that predict protein function. The best of these techniques, Maximum Curiosity, improved the baseline accuracy of 56-77%. This article shows that Active Learning is a useful tool for biomedical research, and provides a case study of interest to others facing similar discovery challenges.