Google DeepMind

23 gain · 25 problem · 0 high-evidence

1. gainHealthModerate evidenceImpact 78

   Structural modeling tools such as AlphaFold and RoseTTAFold enable high-accuracy 3D prediction of proteins encoded by multidrug-resistant bacterial genomes, facilitating functional annotation and accelerating design of new antimicrobial compounds.

   By July 30 2026, a narrative review in Journal of Computer-Aided Molecular Design described the use of structural modeling and artificial intelligence for functional prediction of proteins encoded by multidrug-resistant bacterial genomes. It reported that tools such as AlphaFold and RoseTTAFold have enabled high-accuracy three-dimensional structure prediction, facilitating annotation of hypothetical proteins and identification of conserved domains and catalytic sites.

   Jul 31, 2026

   1 source

   • Journal of Computer-Aided Molecular DesignPeer-reviewed
2. problemHealthModerate evidenceImpact 78

   AI-driven modeling for antimicrobial resistance target discovery continues to face persistent challenges around experimental validation and genomic variability, limiting confirmation of predicted functions.

   By July 30 2026, a narrative review in Journal of Computer-Aided Molecular Design described the use of structural modeling and artificial intelligence for functional prediction of proteins encoded by multidrug-resistant bacterial genomes. It reported that tools such as AlphaFold and RoseTTAFold have enabled high-accuracy three-dimensional structure prediction, facilitating annotation of hypothetical proteins and identification of conserved domains and catalytic sites.

   Jul 31, 2026

   1 source

   • Journal of Computer-Aided Molecular DesignPeer-reviewed
3. gainHealthModerate evidenceImpact 74

   Manus architecture using raw 3D CBCT data detected and correctly diagnosed 95% of jaw lesions in 97 patients, outperforming 2D panoramic inputs.

   A cross-sectional study tested four AI chatbots on 97 anonymized CBCT cases of jaw lesions, comparing performance on reconstructed 2D panoramic views and, for Manus, raw 3D DICOM data. Reports were scored for accuracy, relevance and feasibility, revealing statistically significant differences between systems.

   Aug 5, 2026

   1 source

   • Dentomaxillofacial RadiologyPeer-reviewed
4. gainScienceModerate evidenceImpact 73

   AI-driven enzyme engineering enables rapid, precise design of synthetic synzymes that catalyze non-natural reactions for use in pharmaceuticals, biofuels, and environmental remediation.

   As of December 2025, researchers synthesized AI methods for enzyme engineering, using structure-prediction, generative, and reinforcement learning models combined with high-throughput screening to design and optimize enzymes, including synthetic synzymes for non-natural reactions.

   Jul 20, 2026

   1 source

   • MoleculesPeer-reviewed
5. gainHealthModerate evidenceImpact 72

   Large language models directed patients with musculoskeletal complaints to currently practicing, specialty-appropriate providers in the requested city, with ChatGPT achieving 100% appropriateness in the tested queries.

   Researchers prompted ChatGPT, DeepSeek, and Gemini with standardized musculoskeletal complaints for Lynchburg, VA and Trumbull, CT, and judged whether recommended physicians were currently practicing locally in the relevant specialty and whether phone numbers were correct. By the August 13, 2026 publication date, ChatGPT was appropriate in all 17 recommendations, while Gemini and DeepSeek were appropriate in 43% and 40% of recommendations respectively.

   Aug 14, 2026

   1 source

   • JAAOS: Global Research and ReviewsPeer-reviewed
6. gainScienceHigh evidenceImpact 71

   A three-track neural network that jointly processes sequence, distance, and coordinate information enables accurate prediction of protein structures and protein-protein complexes, approaching DeepMind's accuracy.

   In work published August 20, 2021, researchers built on the CASP14-era DeepMind approach to protein folding. They developed RoseTTAFold, a three-track network that processes sequence, distance, and coordinate information simultaneously, achieving accuracies approaching those of DeepMind.

   Jul 13, 2026

   5 sources

   • SciencePeer-reviewed
   • Annual Review of BiophysicsPeer-reviewed
   • Acta Crystallographica Section D Structural BiologyPeer-reviewed
   • MedComm – Future MedicinePeer-reviewed
   • Quarterly Reviews of BiophysicsPeer-reviewed
7. gainScienceHigh evidenceImpact 71

   AlphaFold provides a computational method that regularly predicts protein three-dimensional structures with atomic accuracy from amino acid sequence alone, even without homologous structures.

   On July 15, 2021, Nature published the AlphaFold study describing a redesigned neural network that predicts the three-dimensional structure a protein will adopt based solely on its amino acid sequence. The authors reported validation in CASP14, where the model regularly achieved atomic accuracy even when no homologous structure was available and performed competitively with experimental structures.

   Jul 13, 2026

   5 sources

   • NaturePeer-reviewed
   • International Journal of Molecular SciencesPeer-reviewed
   • MoleculesPeer-reviewed
   • Nature MethodsPeer-reviewed
   • Genome BiologyPeer-reviewed
8. gainHealthModerate evidenceImpact 69

   AI models Gemini and FaceApp generated post-denture facial images rated as esthetically comparable to actual clinical outcomes by patients and experts.

   A prospective study of 14 edentulous patients compared actual post-denture facial photographs with AI-generated predictions from pretreatment images using Gemini and FaceApp, assessing patient preference, expert esthetic ratings, and quantitative anthropometric measurements.

   Aug 31, 2026

   1 source

   • European Journal of DentistryPeer-reviewed
9. problemHealthModerate evidenceImpact 69

   AI simulations failed to accurately reproduce quantitative facial anthropometric changes after denture placement despite visual similarity.

   A prospective study of 14 edentulous patients compared actual post-denture facial photographs with AI-generated predictions from pretreatment images using Gemini and FaceApp, assessing patient preference, expert esthetic ratings, and quantitative anthropometric measurements.

   Aug 31, 2026

   1 source

   • European Journal of DentistryPeer-reviewed
10. problemHealthModerate evidenceImpact 68

    OpenEvidence and ChatGPT-5 demonstrated greater agreement with the expert reference standard, although clinically significant diagnostic errors were observed across all evaluated systems.

    Purpose The purpose of this study was to evaluate and compare four AI software programs-ChatGPT-5, Microsoft Copilot, Google Gemini (V 2.5), and OpenEvidence-in generating comprehensive dental treatment plans for minimally destructed and severely mutilated teeth using identical clinical inputs. Material and methods Ten anonymized clinical cases, each consisting of 1 intraoral photograph and 1 corresponding periapical radiograph, were independently submitted to each AI software program using a standardized prompt.

    Aug 23, 2026

    1 source

    • Journal of ProsthodonticsPeer-reviewed
11. problemHealthModerate evidenceImpact 68

    When used zero-shot to identify lumbar disc herniations on sagittal MRI, Gemini 3.1 Pro produced low specificity and a substantial false-positive burden, with T1+T2 input performing worse than T1-only.

    Researchers tested Gemini 3.1 Pro in a zero-shot setting to detect lumbar disc herniations on sagittal MRI from 119 SPIDER cases (26% prevalence). Using only the mid-sagittal slice and a forced binary prompt, T1-only achieved 70% accuracy with 58% sensitivity and 74% specificity, while paired T1+T2 achieved 58% accuracy with 77% sensitivity and 51% specificity.

    Aug 14, 2026

    1 source

    • European Spine JournalPeer-reviewed
12. problemHealthModerate evidenceImpact 68

    When recommending musculoskeletal providers, LLMs at times provided inaccurate phone numbers and contact information, potentially preventing patients from reaching the appropriate clinic.

    Researchers prompted ChatGPT, DeepSeek, and Gemini with standardized musculoskeletal complaints for Lynchburg, VA and Trumbull, CT, and judged whether recommended physicians were currently practicing locally in the relevant specialty and whether phone numbers were correct. By the August 13, 2026 publication date, ChatGPT was appropriate in all 17 recommendations, while Gemini and DeepSeek were appropriate in 43% and 40% of recommendations respectively.

    Aug 14, 2026

    1 source

    • JAAOS: Global Research and ReviewsPeer-reviewed