Silent-Speech Brain Control, an Ovarian Precancer Marker and AlphaGenome 2

This briefing covers self-paced silent-speech control for one person with ALS, a candidate blood marker for ovarian precancer, AlphaGenome 2's cell-specific variant predictions, mouthguard evidence on concussion risk and a visible-light photodiode reaching 10 gigabits per second. Each story retains its sample, validation, deployment and manufacturing limitations.

Stories in this briefing

1. Silent Speech Controlled Smart Devices Without Timing Cues

   One participant with ALS used an implanted electrocorticography interface to control smart devices by silently miming commands whenever he chose, without a screen or audio timing cue. Across 18 supervised sessions, median accuracy for a 14-command vocabulary reached 97.1%. The advance is self-paced control, not unrestricted thought reading. It remains a single-participant feasibility study with a constrained vocabulary, and larger home studies must test durability, safety and false activations.

   Read the complete story  →
2. A Blood Marker Detected Microscopic Ovarian Precursors

   Researchers measured the long noncoding RNA Tu-Stroma in stored serum samples from 506 people. In an independent validation set, a locked threshold detected 92.3% of microscopic ovarian-cancer precursor lesions and correctly classified 98.2% of healthy controls. The result is promising because these lesions are usually invisible to imaging, but the retrospective case-control design may overstate real-world separation. Prospective screening studies must establish reproducibility, false-positive rates and whether testing improves clinical outcomes.

   Read the complete story  →
3. AlphaGenome 2 Adds Cellular Context to Variant Prediction

   DeepMind's AlphaGenome 2 predicts how DNA variants affect regulation in specific cell types and states. In the company's TenK10K benchmark, it recovered 30% of tested causal variants, versus 16% for an adapted earlier model. That improvement could help researchers prioritize experiments, but seven in ten causal variants were still missed. The paper is a preprint, central results are developer-led, and predictions require independent and laboratory validation before they can support biological or clinical conclusions.

   Read the complete story  →
4. Mouthguard Data Challenges Older Concussion Risk Estimates

   Mouthguard sensors recorded 3,851 sports impacts, including 47 diagnosed concussions, and found peak linear acceleration predicted injury better than rotational acceleration. The model placed a 50% concussion risk near 100 times gravitational acceleration, below an older estimate based partly on helmet sensors. The result could influence helmet testing, but 47 cases cannot establish one universal threshold. Independent datasets must test the curve across sports, ages, sexes and impact directions.

   Read the complete story  →
5. A Visible-Light Photodiode Received Data at 10 Gigabits per Second

   A tungsten diselenide photodiode integrated with a thin-film lithium niobate waveguide received visible-light data at up to 10 gigabits per second. It also reached 1.12 gigahertz measured bandwidth and 95.6% external quantum efficiency under reported conditions. The device could support sensing, displays and short optical links, but it is a laboratory detector, not a packaged transceiver. Reliability, receiver sensitivity, energy per bit and scalable fabrication remain unproven.

   Read the complete story  →