October 17–18, 2026 | Gene expression, molecular diagnostics, computational medicine and epigenetics
Editorial scope. Verified AI-enabled biological research and noteworthy scientific activity from October 17–18, with a separately dated October 15 epigenetics spotlight. Earlier BunCosmos stories are not repeated.
At a glance
• A blood test teaches AI to recognize how cells respond to Alzheimer’s disease
• AI reads cancer biopsy patterns to examine the risk of recurrence after radiation
• AI protein scientists meet in Beijing to investigate the next generation of biological models
• Spotlight: What makes identical DNA behave differently? Machine learning reveals hidden epigenetic changes
1. A blood test teaches AI to recognize how cells respond to Alzheimer’s disease
OCTOBER 17 PEER-REVIEWED STUDY | UNITED STATES / AUSTRALIA | AI • GENE EXPRESSION • NEURODEGENERATION
Instead of searching only for a familiar Alzheimer's protein in the bloodstream, scientists exposed standardized laboratory neurons to blood plasma and measured how the cells changed their gene activity. A machine-learning classifier then looked for patterns that might distinguish people with early Alzheimer's disease from unaffected participants. The approach treats the cells themselves as sensitive detectors of biological signals carried in blood.
Published on October 17 in Scientific Reports, the exploratory study analyzed 191 retrospectively collected human samples across three cohorts. It tested gene-expression responses using a prespecified 84-gene panel, ultimately identifying a small signature for classification. Accuracy remained limited: the reported area under the curve was 0.64 in an 82-sample validation set and 0.77 in a 23-sample blinded test. These figures do not establish a dependable diagnostic test, particularly given the small blinded sample.
Why it matters: The study shows a different way to use AI in molecular diagnostics: learning from the biological reaction of cells, not merely measuring the concentration of a single marker. Larger prospective studies must establish whether it adds useful information to existing Alzheimer’s tests.
Sources: Scientific Reports — original paper, October 17
2. AI reads cancer biopsy patterns to examine the risk of recurrence after radiation
OCTOBER 17 PEER-REVIEWED STUDY | SWEDEN | AI PATHOLOGY • CANCER PROGNOSIS
Researchers at Karolinska Institutet evaluated whether an AI system could extract useful cancer-risk signals from routine digitized prostate biopsy slides. The model combines tissue-image features with existing clinical measurements and had originally been developed in patients treated by surgery. In its new test, it was applied without retraining to 353 patients who had received radiotherapy.
Published October 17, the study found that AI-derived scores may reveal meaningful differences within established clinical risk groups. However, the model's performance gains over European Association of Urology classifications were not statistically significant. The 353 patients came from the same overarching trial population used to develop the earlier model, and only eight died from prostate cancer during follow-up, limiting certainty about long-term mortality prediction.
Why it matters: AI may help extract subtle signs of tumour behaviour from material hospitals already collect. This is a promising validation step, not a proven basis for changing radiation decisions; independent external and prospective testing are needed.
Sources: Scientific Reports — original study, October 17
3. AI protein scientists meet in Beijing to investigate the next generation of biological models
OCTOBER 17–18 RESEARCH EVENT | CHINA / INTERNATIONAL | PROTEIN DESIGN • BIOLOGICAL AI
The ISCB-China Workshop on AI for Protein Science is taking place in Beijing from October 16 to 18, bringing together researchers working on protein function, molecular design and computational biology. Sessions on October 17–18 include AI-assisted searches for proteins that bind DNA and RNA, single-cell recognition of pathogens, prediction of protein-complex structures, AI-guided protein design and the classification of distant protein families.
The published programme outlines approaches ranging from large protein language models to multimodal enzyme analysis and experimental design. These are scheduled presentations and research abstracts, not independently verified breakthroughs announced during the meeting. The programme nonetheless offers a useful picture of where AI-driven biology is moving, particularly across Chinese universities and research institutes.
Why it matters: The field is expanding beyond predicting a protein’s shape toward testing how molecules interact, identifying functions and choosing experiments that can verify computer-generated hypotheses.
Sources: ISCB-China — official October 16–18 programme • ISCB-China — graduate research forum
4. What makes identical DNA behave differently? Machine learning reveals hidden epigenetic changes
RESEARCH SPOTLIGHT | ORIGINAL CELL PAPER: OCTOBER 15 | UNITED STATES | EPIGENETICS • CHROMATIN
Each cell carries two versions of most chromosomes, one inherited from each parent. Those copies can be regulated differently, even when the DNA sequence in a particular region is identical. Researchers led by scientists at the University of Utah and University of Washington combined long-read DNA measurements with a machine-learning system called FIRE to map accessible parts of chromatin, the packaged form of DNA, on both chromosome copies.
The October 15 Cell paper reported that around 40% of the regulatory differences they detected between chromosome copies had no underlying sequence difference at that specific location. Some patterns persisted as cells divided, consistent with stable epigenetic alterations. The authors also examined a disease-associated regulatory switch near an immune gene and identified previously overlooked candidate regulatory regions. The findings reveal important sources of biological variation, but many of the new regions have not yet been experimentally assigned a function.
Why it matters: AI can help researchers see which DNA regions act as gene-control switches, while epigenetics explains why a genome is not a fixed set of instructions. The study is from October 15 and is included as a research spotlight, not new October 17–18 news.
Sources: Cell — original study, October 15 • University of Utah Health — study explanation
Prepared: October 18, 2026 | BunCosmos AI & Biology edition