AI & Technology - Sci-Tech Bulletin

AI-Powered CAR-T Analysis Could Predict Cancer Immunotherapy Response

A machine learning-based method automatically quantifies CAR-T immunological synapse quality using high-resolution images generated with a glass-supported planar lipid bilayer. Artificial neural networks combine object detection and instance segmentation to identify individual cells and measure F-actin, perforin, tumor-antigen, and pZeta signals. The system could analyze thousands of CAR-T cells and demonstrate differences between samples from a clinical responder and non-responder. The findings support CAR-T immunological synapse quality as a potential composite biomarker and provide a foundation for the proposed Synapse Predicts Efficacy assay for CAR-T development and translational cancer immunotherapy research.

AI-Powered CAR-T Analysis Could Predict Cancer Immunotherapy Response Read More »

An AI-Discovered Alzheimer’s Drug Candidate Targets Galectin-3

An AI-based platform “Llama-Gram” have been used to identify FJMU1887, a novel small-molecule Galectin-3 inhibitor designed for brain penetration. FJMU1887 bound Gal-3 has been found to disrupt Gal-3–TREM2 interactions and reduced inflammatory signaling in microglial cells, including TNF-α inhibition. When studied in 14-month-old 5×FAD mice, 30-day oral treatment reduced Gal-3 expression, microglial activation, neuroinflammation and Aβ pathology while restoring synaptic markers and improving cognitive performance. The findings establish FJMU1887 as a promising preclinical lead, while long-term safety, selectivity and human efficacy remain to be determined.

An AI-Discovered Alzheimer’s Drug Candidate Targets Galectin-3 Read More »

Wireless 60-Channel AI Neurostimulation Platform Targets Real-Time Brain States

A Smart Neurostimulation System is a wireless 60-channel platform that integrates neural sensing, onboard spectral processing, AI classification and programmable stimulation. In three freely moving sheep, movement classifiers achieved cross-validated AUCs of 0.924, 0.979 and 0.969, with performance remaining strong across subsequent weeks. Stimulation experiments demonstrated systematic, parameter-dependent changes in alpha and gamma neural activity, while effects varied considerably by stimulation location. Histological analysis found no adverse tissue response to the tested lead types. The findings establish core technical functionality and a preclinical foundation for personalized, biomarker-guided neuromodulation, while therapeutic efficacy and closed-loop memory benefits remain to be demonstrated.

Wireless 60-Channel AI Neurostimulation Platform Targets Real-Time Brain States Read More »

Autoimmune Disease Prediction Using Deep Learning and TCR Repertoire Analysis

Scientists developed two deep-learning models, AutoY and LSTMY, to predict four autoimmune diseases—type 1 diabetes, rheumatoid arthritis, multiple sclerosis and idiopathic aplastic anemia from T-cell receptor sequencing data. AutoY generally outperformed LSTMY, achieving AUC values of 0.9991 for T1D and 0.9961 for MS, with high accuracy, sensitivity and specificity. LSTMY also performed strongly, particularly for T1D and MS, while showing greater robustness in some evaluations. In future, combining TCR sequencing, deep learning and immune-repertoire analysis could eventually support biomarker discovery, research-oriented screening and precision-medicine applications.

Autoimmune Disease Prediction Using Deep Learning and TCR Repertoire Analysis Read More »

CRISPR-GPT Is Changing AI-Powered Gene Editing

The study demonstrates CRISPR-GPT as an AI co-pilot capable of supporting gene-editing research from experimental planning through analysis. Using a 288-case Gene-editing bench, human-expert evaluations and wet-lab demonstrations, researchers found strong performance across planning, delivery selection, guide-RNA design and scientific Q&A. GPT-4o-powered CRISPR-GPT achieved above 0.99 accuracy, precision, recall and F1 in planning tasks. Its Q&A system outperformed GPT-4o in accuracy, reasoning and conciseness. Two beginner researchers successfully completed AI-guided gene-editing experiments, including multigene knockout and gene activation. The findings suggest potential benefits for research efficiency, reproducibility and accessibility, while emphasizing human oversight, safety and privacy.

CRISPR-GPT Is Changing AI-Powered Gene Editing Read More »

Decoding Cancer Treatment Outcomes with Explainable AI and Multimodal Real-World Data

Explainable artificial intelligence (xAI) combined with multimodal real-world data can improve personalized cancer outcome prediction. Using 350 clinical, laboratory, imaging, treatment, genetic and patient-specific variables from 15,726 patients across 38 solid cancer types, deep-learning models predicted overall survival and time to next treatment more accurately than conventional prognostic scores. xAI identified 114 key markers accounting for 90% of the model’s decision process and uncovered 1,373 prognostic interactions. Important factors included CRP, fT3, ECOG performance status, M stage and LDH. External validation in 3,288 lung cancer patients demonstrated strong reproducibility, supporting potential clinical decision-making applications.

Decoding Cancer Treatment Outcomes with Explainable AI and Multimodal Real-World Data Read More »

An Evidence-Based Digital Platforms Can Help Cancer Patients Make Informed Decisions

A web-based knowledge database can provide cancer patients and patient navigators with evidence-based, easy-to-understand information. Using logfile data and patient surveys, it has been found that 65.9% of eligible patients accessed the platform. The most popular topics searched includes therapy, nutrition, side effects, and carcinogenesis. Nearly 69% of users reported that the database helped them make more informed decisions, while 75.7% found the information they needed. This suggests that accessible, trustworthy digital information can support health literacy and informed decision-making, by further narrowing technical barriers and upgrading the digital skills.

An Evidence-Based Digital Platforms Can Help Cancer Patients Make Informed Decisions Read More »

JPred4: Advanced Protein Secondary Structure Prediction for Modern Bioinformatics

JPred4 is an upgraded protein secondary structure prediction server designed to help researchers analyze protein sequences when experimental structures are unavailable. Its JNet 2.3.1 algorithm combines neural-network prediction, PSI-BLAST-derived evolutionary information, multiple sequence alignments, and improved HMM processing. Blind testing of the server efficiency on 150 previously unseen sequences achieved 82.0% three-state secondary structure accuracy, while solvent accessibility prediction reached up to 90.0%. The platform also offers batch processing, faster analysis, improved visualizations, and comprehensive alignment reports for protein structure modeling, fold recognition, functional-domain identification, mutation planning, and large-scale bioinformatics research.

JPred4: Advanced Protein Secondary Structure Prediction for Modern Bioinformatics Read More »

PEtracer Technology: Revolutionizing Cell Lineage Mapping in Cancer Research

PEtracer is a novel prime editing-based lineage tracing technology that enables simultaneous mapping of cell lineage, gene expression, and spatial organization within intact tissues. By integrating prime editing with single-cell RNA sequencing and MERFISH spatial transcriptomics, cellular phylogenies are accurately reconstructed and tracked tumor evolution in a mouse model of metastatic breast cancer. PEtracer achieves high accuracy in lineage reconstruction while preserving spatial context, making it a powerful tool for studying development, tissue regeneration, cancer progression, and other complex biological processes.

PEtracer Technology: Revolutionizing Cell Lineage Mapping in Cancer Research Read More »

A new Diagnostic Score offers faster diagnosis of primary Polydipsia and AVP deficiency

A simple diagnostic score that helps doctors tell the difference between arginine vasopressin deficiency, a form of diabetes insipidus, and primary polydipsia (excessive water drinking), using routine blood tests, symptoms, and medical history. The tool correctly identified patients with 91% diagnostic accuracy and enabled diagnosis in about 75% of cases without invasive testing. It also reliably distinguished between the two conditions using clear score thresholds with 93% specificity. This practical, evidence-based approach could speed up diagnosis, reduce unnecessary specialist referrals and testing, improve patient comfort, and make better use of healthcare resources.

A new Diagnostic Score offers faster diagnosis of primary Polydipsia and AVP deficiency Read More »