Search for: All records

Alzheimer’s disease (AD) presents significant challenges in clinical practice due to its heterogeneous manifestation and variable progression rates. This work develops a comprehensive anatomical staging framework to predict progression from mild cognitive impairment (MCI) to AD. Using the ADNI database, the scalable Subtype and Stage Inference (s-SuStaIn) model was applied to 118 neuroanatomical features from cognitively normal (n = 504) and AD (n = 346) participants. The framework was validated on 808 MCI participants through associations with clinical progression, CSF and FDG-PET biomarkers, and neuropsychiatric measures, while adjusting for common confounders (age, gender, education, and APOE ε4 alleles). The framework demonstrated superior prognostic accuracy compared to traditional risk assessment (C-index = 0.73 vs. 0.62). Four distinct disease subtypes showed differential progression rates, biomarker profiles (FDG-PET and CSF Aβ42), and cognitive trajectories: Subtype 1, subcortical-first pattern; Subtype 2, executive–cortical pattern; Subtype 3, disconnection pattern; and Subtype 4, frontal–executive pattern. Stage-dependent changes revealed systematic deterioration across diverse cognitive domains, particularly in learning acquisition, visuospatial processing, and functional abilities. This data-driven approach captures clinically meaningful disease heterogeneity and improves prognostication in MCI, potentially enabling more personalized therapeutic strategies and clinical trial design. 

Alzheimer’s disease (AD) is a complex and progressive neurodegenerative condition with significant societal impact. Understanding the temporal dynamics of its pathology is essential for advancing therapeutic interventions. Empirical and anatomical evidence indicates that network decoupling occurs as a result of gray matter atrophy. However, the scarcity of longitudinal clinical data presents challenges for computer-based simulations. To address this, a first-principles-based, physics-constrained Bayesian framework is proposed to model time-dependent connectome dynamics during neurodegeneration. This temporal diffusion network framework segments pathological progression into discrete time windows and optimizes connectome distributions for biomarker Bayesian regression, conceptualized as a learning problem. The framework employs a variational autoencoder-like architecture with computational enhancements to stabilize and improve training efficiency. Experimental evaluations demonstrate that the proposed temporal meta-models outperform traditional static diffusion models. The models were evaluated using both synthetic and real-world MRI and PET clinical datasets that measure amyloid beta, tau, and glucose metabolism. The framework successfully distinguishes normative aging from AD pathology. Findings provide novel support for the “decoupling” hypothesis and reveal eigenvalue-based evidence of pathological destabilization in AD. Future optimization of the model, integrated with real-world clinical data, is expected to improve applications in personalized medicine for AD and other neurodegenerative diseases. 

The overlapping molecular pathophysiology of Alzheimer’s Disease (AD), Amyotrophic Lateral Sclerosis (ALS), and Frontotemporal Dementia (FTD) was analyzed using relationships from a knowledge graph of 33+ million biomedical journal articles. The unsupervised learning rank aggregation algorithm from SemNet 2.0 compared the most important amino acid, peptide, and protein (AAPP) nodes connected to AD, ALS, or FTD. FTD shared 99.9% of its nodes with ALS and AD; AD shared 64.2% of its nodes with FTD and ALS; and ALS shared 68.3% of its nodes with AD and FTD. The results were validated and mapped to functional biological processes using supervised human supervision and an external large language model. The overall percentages of mapped intersecting biological processes were as follows: inflammation and immune response, 19%; synapse and neurotransmission, 19%; cell cycle, 15%; protein aggregation, 12%; membrane regulation, 11%; stress response and regulation, 9%; and gene regulation, 4%. Once normalized for node count, biological mappings for cell cycle regulation and stress response were more prominent in the intersection of AD and FTD. Protein aggregation, gene regulation, and energetics were more prominent in the intersection of ALS and FTD. Synapse and neurotransmission, membrane regulation, and inflammation and immune response were greater at the intersection of AD and ALS. Given the extensive molecular pathophysiology overlap, small differences in regulation, genetic, or environmental factors likely shape the underlying expressed disease phenotype. The results help prioritize testable hypotheses for future clinical or experimental research. 

This work introduces TrialSieve, a novel framework for biomedical information extraction that enhances clinical meta-analysis and drug repurposing. By extending traditional PICO (Patient, Intervention, Comparison, Outcome) methodologies, TrialSieve incorporates hierarchical, treatment group-based graphs, enabling more comprehensive and quantitative comparisons of clinical outcomes. TrialSieve was used to annotate 1609 PubMed abstracts, 170,557 annotations, and 52,638 final spans, incorporating 20 unique annotation categories that capture a diverse range of biomedical entities relevant to systematic reviews and meta-analyses. The performance (accuracy, precision, recall, F1-score) of four natural-language processing (NLP) models (BioLinkBERT, BioBERT, KRISSBERT, PubMedBERT) and the large language model (LLM), GPT-4o, was evaluated using the human-annotated TrialSieve dataset. BioLinkBERT had the best accuracy (0.875) and recall (0.679) for biomedical entity labeling, whereas PubMedBERT had the best precision (0.614) and F1-score (0.639). Error analysis showed that NLP models trained on noisy, human-annotated data can match or, in most cases, surpass human performance. This finding highlights the feasibility of fully automating biomedical information extraction, even when relying on imperfectly annotated datasets. An annotator user study (n = 39) revealed significant (p < 0.05) gains in efficiency and human annotation accuracy with the unique TrialSieve tree-based annotation approach. In summary, TrialSieve provides a foundation to improve automated biomedical information extraction for frontend clinical research. 

Amyotrophic lateral sclerosis (ALS) has an interactive, multifactorial etiology that makes treatment success elusive. This study evaluates how regulatory dynamics impact disease progression and treatment. Computational models of wild-type (WT) and transgenic SOD1-G93A mouse physiology dynamics were built using the first-principles-based first-order feedback framework of dynamic meta-analysis with parameter optimization. Two in silico models were developed: a WT mouse model to simulate normal homeostasis and a SOD1-G93A ALS model to simulate ALS pathology dynamics and their response to in silico treatments. The model simulates functional molecular mechanisms for apoptosis, metal chelation, energetics, excitotoxicity, inflammation, oxidative stress, and proteomics using curated data from published SOD1-G93A mouse experiments. Temporal disease progression measures (rotarod, grip strength, body weight) were used for validation. Results illustrate that untreated SOD1-G93A ALS dynamics cannot maintain homeostasis due to a mathematical oscillating instability as determined by eigenvalue analysis. The onset and magnitude of homeostatic instability corresponded to disease onset and progression. Oscillations were associated with high feedback gain due to hypervigilant regulation. Multiple combination treatments stabilized the SOD1-G93A ALS mouse dynamics to near-normal WT homeostasis. However, treatment timing and effect size were critical to stabilization corresponding to therapeutic success. The dynamics-based approach redefines therapeutic strategies by emphasizing the restoration of homeostasis through precisely timed and stabilizing combination therapies, presenting a promising framework for application to other multifactorial neurodegenerative diseases. 

The ability to translate Generative Adversarial Networks (GANs) and Variational Autoencoders (VAEs) into different modalities and data types is essential to improve Deep Learning (DL) for predictive medicine. This work presents DACMVA, a novel framework to conduct data augmentation in a cross-modal dataset by translating between modalities and oversampling imputations of missing data. DACMVA was inspired by previous work on the alignment of latent spaces in Autoencoders. DACMVA is a DL data augmentation pipeline that improves the performance in a downstream prediction task. The unique DACMVA framework leverages a cross-modal loss to improve the imputation quality and employs training strategies to enable regularized latent spaces. Oversampling of augmented data is integrated into the prediction training. It is empirically demonstrated that the new DACMVA framework is effective in the often-neglected scenario of DL training on tabular data with continuous labels. Specifically, DACMVA is applied towards cancer survival prediction on tabular gene expression data where there is a portion of missing data in a given modality. DACMVA significantly (p << 0.001, one-sided Wilcoxon signed-rank test) outperformed the non-augmented baseline and competing augmentation methods with varying percentages of missing data (4%, 90%, 95% missing). As such, DACMVA provides significant performance improvements, even in very-low-data regimes, over existing state-of-the-art methods, including TDImpute and oversampling alone. 

Background: Amyloid-β plaques (Aβ) 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 Aβ 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 Aβ 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 Aβ: 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 Aβ or 50% “high reduction” of Aβ compared to untreated control. A 25% Aβ reduction achieved a 28% cognitive improvement, and a 50% Aβ 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 Aβ42 to Aβ40. Supervised learning with an 80% –20% train-test split confirmed Aβ reduction was a key feature for predicting MWM escape latency (R2 = 0.8 to 0.95). Conclusions: Results suggest a 25% reduction in Aβ as a meaningful treatment threshold for improving transgenic AD mouse cognition. 

Alzheimer’s disease has a prolonged asymptomatic phase during which pathological changes accumulate before clinical symptoms emerge. This study aimed to stratify the risk of clinical disease to inform future disease-modifying treatments. Cerebrospinal fluid analysis from participants in the Emory Healthy Brain Study was used to classify individuals based on amyloid beta 42 (Aβ42), total tau (tTau) and phosphorylated tau (pTau) levels. Cognitively normal (CN), biomarker-positive (CN)/BM+individuals were identified using a tTau: Aβ42 ratio > 0.24, determined by Gaussian mixture models. CN/BM+ individuals (n = 134) were classified as having asymptomatic Alzheimer’s disease (AsymAD), while CN, biomarker-negative (CN/BM−) individuals served as controls (n = 134). Cognitively symptomatic, biomarker-positive individuals with an Alzheimer’s disease diagnosis confirmed by the Emory Cognitive Neurology Clinic were labelled as Alzheimer’s disease (n = 134). Study groups were matched for age, sex, race and education. Cerebrospinal fluid samples from these matched Emory Healthy Brain Study groups were analysed using targeted proteomics via selected reaction monitoring mass spectrometry. The targeted cerebrospinal fluid panel included 75 peptides from 58 unique proteins. Machine learning approaches identified a subset of eight peptides (ADQDTIR, AQALEQAK, ELQAAQAR, EPVAGDAVPGPK, IASNTQSR, LGADMEDVCGR, VVSSIEQK, YDNSLK) that distinguished between CN/BM− and symptomatic Alzheimer’s disease samples with a binary classifier area under the curve performance of 0.98. Using these eight peptides, Emory Healthy Brain Study AsymAD cases were further stratified into ‘Control-like’ and ‘Alzheimer’s disease-like’ subgroups, representing varying levels of risk for developing clinical disease. The eight peptides were evaluated in an independent dataset from the Alzheimer’s Disease Neuroimaging Initiative, effectively distinguishing CN/BM− from symptomatic Alzheimer’s disease cases (area under the curve = 0.89) and stratifying AsymAD individuals into control-like and Alzheimer’s disease-like subgroups (area under the curve = 0.89). In the absence of matched longitudinal data, an established cross-sectional event-based disease progression model was employed to assess the generalizability of these peptides for risk stratification. In summary, results from two independent modelling methods and datasets demonstrate that the identified eight peptides effectively stratify the risk of progression from asymptomatic to symptomatic Alzheimer’s disease.