Background & Aims The management of pancreatic cysts poses challenges to both patients and their physicians. We investigated whether a combination of molecular markers and clinical information could improve the classification of pancreatic cysts and management of patients. Methods We performed a multi-center, retrospective study of 130 patients with resected pancreatic cystic neoplasms (12 serous cystadenomas, 10 solid-pseudopapillary neoplasms, 12 mucinous cystic neoplasms, and 96 intraductal papillary mucinous neoplasms). Cyst fluid was analyzed to identify subtle mutations in genes known to be mutated in pancreatic cysts (BRAF, CDKN2A, CTNNB1, GNAS, KRAS, NRAS, PIK3CA, RNF43, SMAD4, TP53 and VHL); to identify loss of heterozygozity at CDKN2A, RNF43, SMAD4, TP53, and VHL tumor suppressor loci; and to identify aneuploidy. The analyses were performed using specialized technologies for implementing and interpreting massively parallel sequencing data acquisition. An algorithm was used to select markers that could classify cyst type and grade. The accuracy of the molecular markers were compared with that of clinical markers, and a combination of molecular and clinical markers. Results We identified molecular markers and clinical features that classified cyst type with 90%–100% sensitivity and 92%–98% specificity. The molecular marker panel correctly identified 67 of the 74 patients who did not require surgery, and could therefore reduce the number of unnecessary operations by 91%. Conclusions We identified a panel of molecular markers and clinical features that show promise for the accurate classification of cystic neoplasms of the pancreas and identification of cysts that require surgery.

Humans observe actions performed by others in many different visual and social settings. What features do we extract and attend when we view such complex scenes, and how are they processed in the brain? To answer these questions, we curated two large-scale sets of naturalistic videos of everyday actions and estimated their perceived similarity in two behavioral experiments. We normed and quantified a large range of visual, action-related, and social-affective features across the stimulus sets. Using a cross-validated variance partitioning analysis, we found that social-affective features predicted similarity judgments better than, and independently of, visual and action features in both behavioral experiments. Next, we conducted an electroencephalography experiment, which revealed a sustained correlation between neural responses to videos and their behavioral similarity. Visual, action, and social-affective features predicted neural patterns at early, intermediate, and late stages, respectively, during this behaviorally relevant time window. Together, these findings show that social-affective features are important for perceiving naturalistic actions and are extracted at the final stage of a temporal gradient in the brain.

We here describe a selected reaction monitoring (SRM)-based approach for the discovery and validation of peptide biomarkers for cancer. The first stage of this approach is the direct identification of candidate peptides through comparison of proteolytic peptides derived from the plasma of cancer patients or healthy individuals. Several hundred candidate peptides were identified through this method, providing challenges for choosing and validating the small number of peptides that might prove diagnostically useful. To accomplish this validation, we used 2D chromatography coupled with SRM of candidate peptides. We applied this approach, called sequential analysis of fractionated eluates by SRM (SAFE-SRM), to plasma from cancer patients and discovered two peptides encoded by the peptidyl-prolyl isomerase A (PPIA) gene whose abundance was increased in the plasma of ovarian cancer patients. At optimal thresholds, elevated levels of at least one of these two peptides was detected in 43 (68.3%) of 63 women with ovarian cancer but in none of 50 healthy controls. In addition to providing a potential biomarker for ovarian cancer, this approach is generally applicable to the discovery of peptides characteristic of various disease states.

Our approach identified combinatorial markers for pancreatic cyst classification that had improved performance relative to the individual features they comprise. In principle, this approach can be applied to any clinical dataset comprising dichotomous, categorical, and continuous-valued parameters.

Humans observe actions performed by others in many different visual and social settings. What features do we extract and attend when we view such complex scenes, and how are they processed in the brain? To answer these questions, we curated two large-scale sets of naturalistic videos of everyday actions and estimated their perceived similarity in two behavioral experiments. We normed and quantified a large range of visual, action-related and social-affective features across the stimulus sets. Using a cross-validated variance partitioning analysis, we found that social-affective features predicted similarity judgments better than, and independently of, visual and action features in both behavioral experiments. Next, we conducted an electroencephalography (EEG) experiment, which revealed a sustained correlation between neural responses to videos and their behavioral similarity. Visual, action, and social-affective features predicted neural patterns at early, intermediate and late stages respectively during this behaviorally relevant time window. Together, these findings show that social-affective features are important for perceiving naturalistic actions, and are extracted at the final stage of a temporal gradient in the brain.

How do we mentally organize our memories of life events? Two episodes may be connected because they share a similar location, time period, activity, spatial environment, or social and emotional content. However, we lack an understanding of how each of these dimensions contributes to the perceived similarity of two life memories. We addressed this question with a data-driven approach, eliciting pairs of real-life memories from participants. Participants annotated the social, purposive, spatial, temporal, and emotional characteristics of their memories. We found that the overall similarity of memories was influenced by all of these factors, but to very different extents. Emotional features were the most consistent single predictor of overall memory similarity. Memories with different emotional tone were reliably perceived to be dissimilar, even when they occurred at similar times and places and involved similar people; conversely, memories with a shared emotional tone were perceived as similar even when they occurred at different times and places, and involved different people. A predictive model explained over half of the variance in memory similarity, using only information about (i) the emotional properties of events and (ii) the primary action or purpose of events. Emotional features may make an outsized contribution to event similarity because they provide compact summaries of an event’s goals and self-related outcomes, which are critical information for future planning and decision making. Thus, in order to understand and improve real-world memory function, we must account for the strong influence of emotional and purposive information on memory organization and memory search.SignificanceOur brains enable us to understand and act within the present, informed by previous, related life experience. But how are our life experiences organized so that one event can be related to another? Theories have suggested that we use spatiotemporal, social, causal, purposive, and emotional dimensions to inter-relate our memories; however, these organizing principles are usually studied using impersonal laboratory stimuli. Here, we mapped and modeled the connections between people’s own annotated life memories. We found that life events are linked by a variety of factors, but are predominantly connected in memory by their primary activity and emotional character. This highlights a need for theories of memory organization and retrieval to better account for the role of high-level actions and emotions.

Random Forest (RF) remains one of the most widely used general purpose classification methods. Two recent largescale empirical studies demonstrated it to be the best overall classification method among a variety of methods evaluated. One of its main limitations, however, is that it is restricted to only axis-aligned recursive partitions of the feature space. Consequently, RF is particularly sensitive to the orientation of the data. Several studies have proposed "oblique" decision forest methods to address this limitation. However, these methods either have a time and space complexity significantly greater than RF, are sensitive to unit and scale, or empirically do not perform as well as RF on real data. One promising oblique method that was proposed alongside the canonical RF method, called Forest-RC (F-RC), has not received as much attention by the community. Despite it being just as old as RF, virtually no studies exist investigating its theoretical or empirical performance. In this work, we demonstrate that F-RC empirically outperforms RF and another recently proposed oblique method called Random Rotation Random Forest, while approximately maintaining the same computational complexity. Furthermore, a variant of F-RC which rank transforms the data prior to learning is especially invariant to affine transformations and robust to data corruption. Open source code is available.

Machine learning has an emerging critical role in high-performance computing to modulate simulations, extract knowledge from massive data, and replace numerical models with efficient approximations. Decision forests are a critical tool because they provide insight into model operation that is critical to interpreting learned results. While decision forests are trivially parallelizable, the traversals of tree data structures incur many random memory accesses and are very slow.We present memory packing techniques that reorganize learned forests to minimize cache misses during classification. The resulting layout is hierarchical. At low levels, we pack the nodes of multiple trees into contiguous memory blocks so that each memory access fetches data for multiple trees. At higher levels, we use leaf cardinality to identify the most popular paths through a tree and collocate those paths in cache lines. We extend this layout with out-of-order execution and cache-line prefetching to increase memory throughput.Together, these optimizations increase the performance of classification in ensembles by a factor of four over an optimized C++ implementation and a factor of 50 over a popular Rlanguage implementation.