A large-scale neural network training framework for generalized estimation of single-trial population dynamics.

Achieving state-of-the-art performance with deep neural population dynamics models requires extensive hyperparameter tuning for each dataset. AutoLFADS is a model-tuning framework that automatically produces high-performing autoencoding models on data from a variety of brain areas and tasks, without behavioral or task information. We demonstrate its broad applicability on several rhesus macaque datasets: from motor cortex during free-paced reaching, somatosensory cortex during reaching with perturbations, and dorsomedial frontal cortex during a cognitive timing task.

Canalization of the Polygenic Risk for Common Diseases and Traits in the UK Biobank Cohort.

Since organisms develop and thrive in the face of constant perturbations due to environmental and genetic variation, species may evolve resilient genetic architectures. We sought evidence for this process, known as canalization, through a comparison of the prevalence of phenotypes as a function of the polygenic score (PGS) across environments in the UK Biobank cohort study. Contrasting seven diseases and three categorical phenotypes with respect to 151 exposures in 408,925 people, the deviation between the prevalence-risk curves was observed to increase monotonically with the PGS percentile in one-fifth of the comparisons, suggesting extensive PGS-by-Environment (PGS×E) interaction. After adjustment for the dependency of allelic effect sizes on increased prevalence in the perturbing environment, cases where polygenic influences are greater or lesser than expected are seen to be particularly pervasive for educational attainment, obesity, and metabolic condition type-2 diabetes. Inflammatory bowel disease analysis shows fewer interactions but confirms that smoking and some aspects of diet influence risk. Notably, body mass index has more evidence for decanalization (increased genetic influence at the extremes of polygenic risk), whereas the waist-to-hip ratio shows canalization, reflecting different evolutionary pressures on the architectures of these weight-related traits. An additional 10 % of comparisons showed evidence for an additive shift of prevalence independent of PGS between exposures. These results provide the first widespread evidence for canalization protecting against disease in humans and have implications for personalized medicine as well as understanding the evolution of complex traits. The findings can be explored through an R shiny app at https://canalization-gibsonlab.shinyapps.io/rshiny/.

What Threshold of Amyloid Reduction Is Necessary to Meaningfully Improve Cognitive Function in Transgenic Alzheimer's Disease Mice?

Background: Amyloid-ß plaques (Aß) are associated with Alzheimer's disease (AD). Pooled assessment of amyloid reduction in transgenic AD mice is critical for expediting anti-amyloid AD therapeutic research. Objective: The mean threshold of Aß reduction necessary to achieve cognitive improvement was measured via pooled assessment (n = 594 mice) of Morris water maze (MWM) escape latency of transgenic AD mice treated with substances intended to reduce Aß via reduction of beta-secretase cleaving enzyme (BACE). Methods: Machine learning and statistical methods identified necessary amyloid reduction levels using mouse data (e.g., APP/PS1, LPS, Tg2576, 3xTg-AD, control, wild type, treated, untreated) curated from 22 published studies. Results: K-means clustering identified 4 clusters that primarily corresponded with level of Aß: untreated transgenic AD control mice, wild type mice, and two clusters of transgenic AD mice treated with BACE inhibitors that had either an average 25% "medium reduction" of Aß or 50% "high reduction" of Aß compared to untreated control. A 25% Aß reduction achieved a 28% cognitive improvement, and a 50% Aß reduction resulted in a significant 32% improvement compared to untreated transgenic mice (p < 0.05). Comparatively, wild type mice had a mean 41% MWM latency improvement over untreated transgenic mice (p < 0.05). BACE reduction had a lesser impact on the ratio of Aß42 to Aß40. Supervised learning with an 80% -20% train-test split confirmed Aß reduction was a key feature for predicting MWM escape latency (R2 = 0.8 to 0.95). Conclusions: Results suggest a 25% reduction in Aß as a meaningful treatment threshold for improving transgenic AD mouse cognition.

sEBM: Scaling Event Based Models to Predict Disease Progression via Implicit Biomarker Selection and Clustering.

The Event Based Model (EBM) is a probabilistic generative model to explore biomarker changes occurring as a disease progresses. Disease progression is hypothesized to occur through a sequence of biomarker dysregulation "events". The EBM estimates the biomarker dysregulation event sequence. It computes the data likelihood for a given dysregulation sequence, and subsequently evaluates the posterior distribution on the dysregulation sequence. Since the posterior distribution is intractable, Markov Chain Monte-Carlo is employed to generate samples under the posterior distribution. However, the set of possible sequences increases as N! where N is the number of biomarkers (data dimension) and quickly becomes prohibitively large for effective sampling via MCMC. This work proposes the "scaled EBM" (sEBM) to enable event based modeling on large biomarker sets (e.g. high-dimensional data). First, sEBM implicitly selects a subset of biomarkers useful for modeling disease progression and infers the event sequence only for that subset. Second, sEBM clusters biomarkers with similar positions in the event sequence and only orders the "clusters", with each successive cluster corresponding to the next stage in disease progression. These two modifications used to construct the sEBM method provably reduces the possible space of event sequences by multiple orders of magnitude. The novel modifications are supported by theory and experiments on synthetic and real clinical data provides validation for sEBM to work in higher dimensional settings. Results on synthetic data with known ground truth shows that sEBM outperforms previous EBM variants as data dimensions increase. sEBM was successfully implemented with up to 300 biomarkers, which is a 6-fold increase over previous EBM applications. A real-world clinical application of sEBM is performed using 119 neuroimaging markers from publicly available Alzheimer's Disease Neuroimaging Initiative (ADNI) data to stratify subjects into 6 stages of disease progression. Subjects included cognitively normal (CN), mild cognitive impairment (MCI), and Alzheimer's Disease (AD). sEBM stage is differentiated for the 3 groups (χ2p-value<4.6e-32). Increased sEBM stage is a strong predictor of conversion risk to AD (p-value<2.3e-14) for MCI subjects, as verified with a Cox proportional-hazards model adjusted for age, sex, education and APOE4 status. Like EBM, sEBM does not rely on apriori defined diagnostic labels and only uses cross-sectional data.

Machine Learning Selection of Most Predictive Brain Proteins Suggests Role of Sugar Metabolism in Alzheimer's Disease.

BACKGROUND: The complex and not yet fully understood etiology of Alzheimer's disease (AD) shows important proteopathic signs which are unlikely to be linked to a single protein. However, protein subsets from deep proteomic datasets can be useful in stratifying patient risk, identifying stage dependent disease markers, and suggesting possible disease mechanisms. OBJECTIVE: The objective was to identify protein subsets that best classify subjects into control, asymptomatic Alzheimer's disease (AsymAD), and AD. METHODS: Data comprised 6 cohorts; 620 subjects; 3,334 proteins. Brain tissue-derived predictive protein subsets for classifying AD, AsymAD, or control were identified and validated with label-free quantification and machine learning. RESULTS: A 29-protein subset accurately classified AD (AUC = 0.94). However, an 88-protein subset best predicted AsymAD (AUC = 0.92) or Control (AUC = 0.92) from AD (AUC = 0.98). AD versus Control: APP, DHX15, NRXN1, PBXIP1, RABEP1, STOM, and VGF. AD versus AsymAD: ALDH1A1, BDH2, C4A, FABP7, GABBR2, GNAI3, PBXIP1, and PRKAR1B. AsymAD versus Control: APP, C4A, DMXL1, EXOC2, PITPNB, RABEP1, and VGF. Additional predictors: DNAJA3, PTBP2, SLC30A9, VAT1L, CROCC, PNP, SNCB, ENPP6, HAPLN2, PSMD4, and CMAS. CONCLUSION: Biomarkers were dynamically separable across disease stages. Predictive proteins were significantly enriched to sugar metabolism.

Predictors of Cognitive Decline in Healthy Middle-Aged Individuals with Asymptomatic Alzheimer's Disease.

Alzheimer's disease (AD) progresses through a lengthy asymptomatic period during which pathological changes accumulate prior to development of clinical symptoms. As disease-modifying treatments are developed, tools to stratify risk of clinical disease will be required to guide their use. In this study, we examine the relationship of AD biomarkers in healthy middle-aged individuals to health history, family history, and neuropsychological measures and identify cerebrospinal fluid (CSF) biomarkers to stratify risk of progression from asymptomatic to symptomatic AD. CSF from cognitively normal (CN) individuals (N=1149) in the Emory Healthy Brain Study were assayed for Aß42, total Tau (tTau), and phospho181-Tau (pTau), and a subset of 134 cognitively normal, but biomarker-positive, individuals were identified with asymptomatic AD (AsymAD) based on a locally-determined cutoff value for ratio of tTau to Aß42. These AsymAD cases were matched for demographic features with 134 biomarker-negative controls (CN/BM-) and compared for differences in medical comorbidities and family history. Dyslipidemia emerged as a distinguishing feature between AsymAD and CN/BM-groups with significant association with personal and family history of dyslipidemia. A weaker relationship was seen with diabetes, but there was no association with hypertension. Examination of the full cohort by median regression revealed a significant relationship of CSF Aß42 (but not tTau or pTau) with dyslipidemia and diabetes. On neuropsychological tests, CSF Aß42 was not correlated with performance on any measures, but tTau and pTau were strongly correlated with visuospatial perception and visual episodic memory. In addition to traditional CSF AD biomarkers, a panel of AD biomarker peptides derived from integrating brain and CSF proteomes were evaluated using machine learning strategies to identify a set of 8 peptides that accurately classified CN/BM- and symptomatic AD CSF samples with AUC of 0.982. Using these 8 peptides in a low dimensional t-distributed Stochastic Neighbor Embedding analysis and k-Nearest Neighbor (k=5) algorithm, AsymAD cases were stratified into "Control-like" and "AD-like" subgroups based on their proximity to CN/BM- or AD CSF profiles. Independent analysis of these cases using a Joint Mutual Information algorithm selected a set of 5 peptides with 81% accuracy in stratifying cases into AD-like and Control-like subgroups. Performance of both sets of peptides was evaluated and validated in an independent data set from the Alzheimer's Disease Neuroimaging Initiative. Based on our findings, we conclude that there is an important role of lipid metabolism in asymptomatic stages of AD. Visuospatial perception and visual episodic memory may be more sensitive than language-based abilities to earliest stages of cognitive decline in AD. Finally, candidate CSF peptides show promise as next generation biomarkers for predicting progression from asymptomatic to symptomatic stages of AD.

A deep learning framework for inference of single-trial neural population dynamics from calcium imaging with subframe temporal resolution.

In many areas of the brain, neural populations act as a coordinated network whose state is tied to behavior on a millisecond timescale. Two-photon (2p) calcium imaging is a powerful tool to probe such network-scale phenomena. However, estimating the network state and dynamics from 2p measurements has proven challenging because of noise, inherent nonlinearities and limitations on temporal resolution. Here we describe Recurrent Autoencoder for Discovering Imaged Calcium Latents (RADICaL), a deep learning method to overcome these limitations at the population level. RADICaL extends methods that exploit dynamics in spiking activity for application to deconvolved calcium signals, whose statistics and temporal dynamics are quite distinct from electrophysiologically recorded spikes. It incorporates a new network training strategy that capitalizes on the timing of 2p sampling to recover network dynamics with high temporal precision. In synthetic tests, RADICaL infers the network state more accurately than previous methods, particularly for high-frequency components. In 2p recordings from sensorimotor areas in mice performing a forelimb reach task, RADICaL infers network state with close correspondence to single-trial variations in behavior and maintains high-quality inference even when neuronal populations are substantially reduced.