The human striatum has been implicated in processing reward-related information. More recently, activity in the striatum, particularly the caudate nucleus, has been observed when a contingency between behavior and reward exists, suggesting a role for the caudate in reinforcement-based learning. Using a gambling paradigm, in which affective feedback (reward and punishment) followed simple, random guesses on a trial by trial basis, we sought to investigate the role of the caudate nucleus as reward-related learning progressed. Participants were instructed to make a guess regarding the value of a presented card (if the value of the card was higher or lower than 5). They were told that five different cues would be presented prior to making a guess, and that each cue indicated the probability that the card would be high or low. The goal was to learn the contingencies and maximize the reward attained. Accuracy, as measured by participant's choices, improved throughout the experiment for cues that strongly predicted reward, while no change was observed for unpredictable cues. Event-related fMRI revealed that activity in the caudate nucleus was more robust during the early phases of learning, irrespective of contingencies, suggesting involvement of this region during the initial stages of trial and error learning. Further, the reward feedback signal in the caudate nucleus for well-learned cues decreased as learning progressed, suggesting an evolving adaptation of reward feedback expectancy as a behavior-outcome contingency becomes more predictable.

Predicting danger from previously associated sensory stimuli is essential for survival. Contributions from altered peripheral sensory inputs are implicated in this process, but the underlying mechanisms remain elusive. Here, we use the mammalian olfactory system to investigate such mechanisms. Primary olfactory sensory neurons (OSNs) project their axons directly to the olfactory bulb (OB) glomeruli, where their synaptic release is subject to local and cortical influence and neuromodulation. Pairing optogenetic activation of a single glomerulus with foot shock in mice induces freezing to light stimulation alone during fear retrieval. This is accompanied by an increase in OSN release probability and a reduction in GABAB receptor expression in the conditioned glomerulus. Furthermore, freezing time is positively correlated with the release probability of OSNs in fear-conditioned mice. These results suggest that aversive learning increases peripheral olfactory inputs at the first synapse, which may contribute to the behavioral outcome.

The exploration/exploitation tradeoff - pursuing a known reward vs. sampling from lesser known options in the hope of finding a better payoff - is a fundamental aspect of learning and decision making. In humans, this has been studied using multi-armed bandit tasks. The same processes have also been studied using simplified probabilistic reversal learning (PRL) tasks with binary choices. Our investigations suggest that protocols previously used to explore PRL in mice may prove beyond their cognitive capacities, with animals performing at a no-better-than-chance level. We sought a novel probabilistic learning task to improve behavioral responding in mice, whilst allowing the investigation of the exploration/exploitation tradeoff in decision making. To achieve this, we developed a two-lever operant chamber task with levers corresponding to different probabilities (high/low) of receiving a saccharin reward, reversing the reward contingencies associated with levers once animals reached a threshold of 80% responding at the high rewarding lever. We found that, unlike in existing PRL tasks, mice are able to learn and behave near optimally with 80% high/20% low reward probabilities. Altering the reward contingencies towards equality showed that some mice displayed preference for the high rewarding lever with probabilities as close as 60% high/40% low. Additionally, we show that animal choice behavior can be effectively modelled using reinforcement learning (RL) models incorporating learning rates for positive and negative prediction error, a perseveration parameter, and a noise parameter. This new decision task, coupled with RL analyses, advances access to investigate the neuroscience of the exploration/exploitation tradeoff in decision making.

Previous studies have suggested that negative feedback is more effective in driving learning than positive feedback. We investigated the effect on learning of providing varying amounts of negative and positive feedback while listeners attempted to discriminate between three identical tones; an impossible task that nevertheless produces robust learning. Four feedback conditions were compared during training: 90% positive feedback or 10% negative feedback informed the participants that they were doing equally well, while 10% positive or 90% negative feedback informed them they were doing equally badly. In all conditions the feedback was random in relation to the listeners' responses (because the task was to discriminate three identical tones), yet both the valence (negative vs. positive) and the probability of feedback (10% vs. 90%) affected learning. Feedback that informed listeners they were doing badly resulted in better post-training performance than feedback that informed them they were doing well, independent of valence. In addition, positive feedback during training resulted in better post-training performance than negative feedback, but only positive feedback indicating listeners were doing badly on the task resulted in learning. As we have previously speculated, feedback that better reflected the difficulty of the task was more effective in driving learning than feedback that suggested performance was better than it should have been given perceived task difficulty. But contrary to expectations, positive feedback was more effective than negative feedback in driving learning. Feedback thus had two separable effects on learning: feedback valence affected motivation on a subjectively difficult task, and learning occurred only when feedback probability reflected the subjective difficulty. To optimize learning, training programs need to take into consideration both feedback valence and probability.

Deep-learning-based segmentation models implicitly learn to predict the presence of a structure based on its overall prominence in the training dataset. This phenomenon is observed and accounted for in deep-learning applications such as natural language processing but is often neglected in segmentation literature. The purpose of this work is to demonstrate the significance of class imbalance in deep-learning-based segmentation and recommend tuning of the neural network optimization objective.

Reconstruction of host-pathogen protein interaction networks is of great significance to reveal the underlying microbic pathogenesis. However, the current experimentally-derived networks are generally small and should be augmented by computational methods for less-biased biological inference. From the point of view of computational modelling, data scarcity, data unavailability and negative data sampling are the three major problems for host-pathogen protein interaction networks reconstruction. In this work, we are motivated to address the three concerns and propose a probability weighted ensemble transfer learning model for HIV-human protein interaction prediction (PWEN-TLM), where support vector machine (SVM) is adopted as the individual classifier of the ensemble model. In the model, data scarcity and data unavailability are tackled by homolog knowledge transfer. The importance of homolog knowledge is measured by the ROC-AUC metric of the individual classifiers, whose outputs are probability weighted to yield the final decision. In addition, we further validate the assumption that only the homolog knowledge is sufficient to train a satisfactory model for host-pathogen protein interaction prediction. Thus the model is more robust against data unavailability with less demanding data constraint. As regards with negative data construction, experiments show that exclusiveness of subcellular co-localized proteins is unbiased and more reliable than random sampling. Last, we conduct analysis of overlapped predictions between our model and the existing models, and apply the model to novel host-pathogen PPIs recognition for further biological research.

The emergence of high-throughput RNA-seq data has offered unprecedented opportunities for cancer diagnosis. However, capturing biological data with highly nonlinear and complex associations by most existing approaches for cancer diagnosis has been challenging. In this study, we propose a novel hierarchical feature selection and second learning probability error ensemble model (named HFS-SLPEE) for precision cancer diagnosis. Specifically, we first integrated protein-coding gene expression profiles, non-coding RNA expression profiles, and DNA methylation data to provide rich information; afterward, we designed a novel hierarchical feature selection method, which takes the CpG-gene biological associations into account and can select a compact set of superior features; next, we used four individual classifiers with significant differences and apparent complementary to build the heterogeneous classifiers; lastly, we developed a second learning probability error ensemble model called SLPEE to thoroughly learn the new data consisting of classifiers-predicted class probability values and the actual label, further realizing the self-correction of the diagnosis errors. Benchmarking comparisons on TCGA showed that HFS-SLPEE performs better than the state-of-the-art approaches. Moreover, we analyzed in-depth 10 groups of selected features and found several novel HFS-SLPEE-predicted epigenomics and epigenetics biomarkers for breast invasive carcinoma (BRCA) (e.g., TSLP and ADAMTS9-AS2), lung adenocarcinoma (LUAD) (e.g., HBA1 and CTB-43E15.1), and kidney renal clear cell carcinoma (KIRC) (e.g., IRX2 and BMPR1B-AS1).

The purpose of this study is to investigate whether machine learning with dosiomic, radiomic, and demographic features allows for xerostomia risk assessment more precise than normal tissue complication probability (NTCP) models based on the mean radiation dose to parotid glands.

Clinicians use probability estimates to make a diagnosis. Teaching students to make more accurate probability estimates could improve the diagnostic process and, ultimately, the quality of medical care.

The epidemic increase in the incidence of Human Papilloma Virus (HPV) related Oropharyngeal Squamous Cell Carcinomas (OPSCCs) in several countries worldwide represents a significant public health concern. Although gender neutral HPV vaccination programmes are expected to cause a reduction in the incidence rates of OPSCCs, these effects will not be evident in the foreseeable future. Secondary prevention strategies are currently not feasible due to an incomplete understanding of the natural history of oral HPV infections in OPSCCs. The key parameters that govern natural history models remain largely ill-defined for HPV related OPSCCs and cannot be easily inferred from experimental data. Mathematical models have been used to estimate some of these ill-defined parameters in cervical cancer, another HPV related cancer leading to successful implementation of cancer prevention strategies. We outline a "double-Bayesian" mathematical modelling approach, whereby, a Bayesian machine learning model first estimates the probability of an individual having an oral HPV infection, given OPSCC and other covariate information. The model is then inverted using Bayes' theorem to reverse the probability relationship. We use data from the Surveillance, Epidemiology, and End Results (SEER) cancer registry, SEER Head and Neck with HPV Database and the National Health and Nutrition Examination Surveys (NHANES), representing the adult population in the United States to derive our model. The model contains 8,106 OPSCC patients of which 73.0% had an oral HPV infection. When stratified by age, sex, marital status and race/ethnicity, the model estimated a higher conditional probability for developing OPSCCs given an oral HPV infection in non-Hispanic White males and females compared to other races/ethnicities. The proposed Bayesian model represents a proof-of-concept of a natural history model of HPV driven OPSCCs and outlines a strategy for estimating the conditional probability of an individual's risk of developing OPSCC following an oral HPV infection.

Preclinical magnetic resonance imaging (MRI) is a critical component in a co-clinical research pipeline. Importantly, segmentation of tumors in MRI is a necessary step in tumor phenotyping and assessment of response to therapy. However, manual segmentation is time-intensive and suffers from inter- and intra- observer variability and lack of reproducibility. This study aimed to develop an automated pipeline for accurate localization and delineation of TNBC PDX tumors from preclinical T1w and T2w MR images using a deep learning (DL) algorithm and to assess the sensitivity of radiomic features to tumor boundaries. We tested five network architectures including U-Net, dense U-Net, Res-Net, recurrent residual UNet (R2UNet), and dense R2U-Net (D-R2UNet), which were compared against manual delineation by experts. To mitigate bias among multiple experts, the simultaneous truth and performance level estimation (STAPLE) algorithm was applied to create consensus maps. Performance metrics (F1-Score, recall, precision, and AUC) were used to assess the performance of the networks. Multi-contrast D-R2UNet performed best with F1-score = 0.948; however, all networks scored within 1-3% of each other. Radiomic features extracted from D-R2UNet were highly corelated to STAPLE-derived features with 67.13% of T1w and 53.15% of T2w exhibiting correlation ρ ≥ 0.9 (p ≤ 0.05). D-R2UNet-extracted features exhibited better reproducibility relative to STAPLE with 86.71% of T1w and 69.93% of T2w features found to be highly reproducible (CCC ≥ 0.9, p ≤ 0.05). Finally, 39.16% T1w and 13.9% T2w features were identified as insensitive to tumor boundary perturbations (Spearman correlation (-0.4 ≤ ρ ≤ 0.4). We developed a highly reproducible DL algorithm to circumvent manual segmentation of T1w and T2w MR images and identified sensitivity of radiomic features to tumor boundaries.

Low birth weight is associated with increased risks of health problems in infancy and later life. Among the epidemiological analyses suggesting an association between air pollution and birth weight, few have estimated the effects of black carbon (BC) or together with nitrogen dioxide (NO2), and even fewer studies have used causal modelling.

Traditional methods for investigating work hours rely on an employee's physical presence at the worksite. However, accurately identifying break times at the worksite and distinguishing remote work outside the worksite poses challenges in work hour estimations. Machine learning has the potential to differentiate between human-smartphone interactions at work and off work.

The effectiveness of rotator cuff tear repair surgery is influenced by multiple patient-related, pathology-centred and technical factors, which is thought to contribute to the reported retear rates between 17% and 94%. Adequate patient selection is thought to be essential in reaching satisfactory results. However, no clear consensus has been reached on which factors are most predictive of successful surgery. A clinical decision tool that encompassed all aspects is still to be made. Artificial intelligence (AI) and machine learning algorithms use complex self-learning models that can be used to make patient-specific decision-making tools. The aim of this study is to develop and train an algorithm that can be used as an online available clinical prediction tool, to predict the risk of retear in patients undergoing rotator cuff repair.

In data science, determining proximity between observations is critical to many downstream analyses such as clustering, classification, and prediction. However, when the data's underlying probability distribution is unclear, the function used to compute similarity between data points is often arbitrarily chosen. Here, we present a novel definition of proximity, Semblance, that uses the empirical distribution of a feature to inform the pair-wise similarity between observations. The advantage of Semblance lies in its distribution-free formulation and its ability to place greater emphasis on proximity between observation pairs that fall at the outskirts of the data distribution, as opposed to those toward the center. Semblance is a valid Mercer kernel, allowing its principled use in kernel-based learning algorithms, and for any data modality. We demonstrate its consistently improved performance against conventional methods through simulations and real case studies from diverse applications in single-cell transcriptomics, image reconstruction, and financial forecasting.

Despite technological advancements in recent years, glenoid component loosening remains a common complication after anatomical total shoulder arthroplasty (ATSA) and is one of the main causes of revision surgery. Increasing emphasis is placed on the prevention of glenoid component failure. Previous studies have successfully predicted range of motion, patient-reported outcomes and short-term complications after ATSA using machine learning methods, but an accurate predictive model for (glenoid component) revision is currently lacking. This study aims to use a large international database to accurately predict aseptic loosening of the glenoid component after ATSA using machine learning algorithms.

Humans and other primates share many decision biases, among them our subjective distortion of objective probabilities. When making choices between uncertain rewards we typically treat probabilities nonlinearly: overvaluing low probabilities of reward and undervaluing high ones. A growing body of evidence, however, points to a more flexible pattern of distortion than the classical inverse-S one, highlighting the effect of experimental conditions in shifting the weight assigned to probabilities, such as task feedback, learning, and attention. Here we investigated the role of sequence structure (the order in which gambles are presented in a choice task) in shaping the probability distortion patterns of rhesus macaques: we presented 2 male monkeys with binary choice sequences of MIXED or REPEATED gambles against safe rewards. Parametric modeling revealed that choices in each sequence type were guided by significantly different patterns of probability distortion: whereas we elicited the classical inverse-S-shaped probability distortion in pseudorandomly MIXED trial sequences of gamble-safe choices, we found the opposite pattern consisting of S-shaped distortion, with REPEATED sequences. We extended these results to binary choices between two gambles, without a safe option, and confirmed the unique influence of the sequence structure in which the animals make choices. Finally, we showed that the value of gambles experienced in the past had a significant impact on the subjective value of future ones, shaping probability distortion on a trial-by-trial basis. Together, our results suggest that differences in choice sequence are sufficient to reverse the direction of probability distortion.SIGNIFICANCE STATEMENT Our lives are peppered with uncertain, probabilistic choices. Recent studies showed how such probabilities are subjectively distorted. In the present study, we show that probability distortions in macaque monkeys differ significantly between sequences in which single gambles are repeated (S-shaped distortion), as opposed to being pseudorandomly intermixed with other gambles (inverse-S-shaped distortion). Our findings challenge the idea of fixed probability distortions resulting from inflexible computations, and points to a more instantaneous evaluation of probabilistic information. Past trial outcomes appeared to drive the "gap" between probability distortions in different conditions. Our data suggest that, as in most adaptive systems, probability values are slowly but constantly updated from prior experience, driving measures of probability distortion to either side of the S/inverse-S debate.

Probable stimuli are more often and more quickly detected. While stimulus probability is known to affect decision-making, it can also be explained as a perceptual phenomenon. Using spatial gratings, we have previously shown that probable orientations are also more precisely estimated, even while participants remained naive to the manipulation. We conducted an electrophysiological study to investigate the effect that probability has on perception and visual-evoked potentials. In line with previous studies on oddballs and stimulus prevalence, low-probability orientations were associated with a greater late positive 'P300' component which might be related to either surprise or decision-making. However, the early 'C1' component, thought to reflect V1 processing, was dampened for high-probability orientations while later P1 and N1 components were unaffected. Exploratory analyses revealed a participant-level correlation between C1 and P300 amplitudes, suggesting a link between perceptual processing and decision-making. We discuss how these probability effects could be indicative of sharpening of neurons preferring the probable orientations, due either to perceptual learning, or to feature-based attention.

Previous studies have shown that abrupt onsets randomly appearing at different locations can be ignored with practice, a result that was interpreted as an instance of habituation. Here we addressed whether habituation of capture can be spatially selective and determined by the rate of onset occurrence at different locations, and whether habituation is achieved via spatial suppression applied at the distractor location. In agreement with the habituation hypothesis, we found that capture attenuation was larger where the onset distractor occurred more frequently, similarly to what has been documented for feature-singleton distractors (the "distractor-location effect"), and that onset interference decreased across trials at both the high- and low-probability distractor locations. By contrast, evidence was inconclusive as to whether distractor filtering was also accompanied by a larger impairment in target processing when it appeared at the more likely distractor location (the "target-location effect"), as instead previously reported for feature-singleton distractors. Finally, here we discuss how and to what extent distractor rejection based on statistical learning and habituation of capture are different, and conclude that the two notions are intimately related, as the Sokolov model of habituation operates by comparing the upcoming sensory input with expectation based on the statistics of previous stimulation.

Faced with potential harm, individuals must estimate the probability of threat and initiate an appropriate fear response. In the prevailing view, threat probability estimates are relayed to the ventrolateral periaqueductal gray (vlPAG) to organize fear output. A straightforward prediction is that vlPAG single-unit activity reflects fear output, invariant of threat probability. We recorded vlPAG single-unit activity in male, Long Evans rats undergoing fear discrimination. Three 10 s auditory cues predicted unique foot shock probabilities: danger (p=1.00), uncertainty (p=0.375) and safety (p=0.00). Fear output was measured by suppression of reward seeking over the entire cue and in one-second cue intervals. Cued fear non-linearly scaled to threat probability and cue-responsive vlPAG single-units scaled their firing on one of two timescales: at onset or ramping toward shock delivery. VlPAG onset activity reflected threat probability, invariant of fear output, while ramping activity reflected both signals with threat probability prioritized.