This article was generated entirely by AI without human authorship. Please read with discretion, or have an AI verify its accuracy.
AI TranslationSimplified ChineseEnglish
Summary:
Nine papers added this week spanning structural biology, AI drug discovery, and neuroscience, each with a curated summary and research highlight.
This issue rounds up nine papers added to my personal Zotero library between 31 August and 6 September 2026, listed in reverse order of accession. All dates use Beijing time and refer to when an item was added; each paper’s publication date is given separately.
The notes below were assembled by AI from the abstracts; the Biomni entry includes an extended summary. Full texts were not read. Overviews and Highlights are meant for quick scanning — please consult the original papers for methods and findings.
01 · Phage therapy: from basic biology to clinical application
Nature Reviews Microbiology · Published 2026-08-27
Holly A. Sinclair, Ruby C. Y. Lin, Nouri L. Ben Zakour, Paul L. Bollyky, Jonathan R. Iredell
Bacteriophages have been used to treat bacterial infections in humans for over a century, yet in vitro results still predict clinical outcomes poorly and no consensus has formed on how best to use them. This review starts from phage biology to discuss how to choose phages rationally and understand the interplay among bacteria, phages, and the human host. The authors stress that the adaptations bacteria and phages undergo in vivo must be built into treatment design, and that paradigms borrowed from conventional antibiotics cannot simply be copied over. They set out clinical strategies for different infections with their strengths and limits, and also discuss obstacles in manufacturing, scaling, and regulation. How phages interact with the human immune system is another key determinant of whether treatment succeeds. The review casts phages as an important avenue against antimicrobial resistance, but one whose use depends on deeper mechanistic understanding and case-specific choices.
Highlight · Phage therapy has to reckon with the dynamic relationship between bacteria, phages, and the host; in vitro activity cannot be equated with clinical efficacy.
02 · The evolving role of structural biology in pharma: integration of X-ray crystallography, cryo-electron microscopy and beyond
Acta Crystallographica Section D: Structural Biology · Published 2026-06-01
J. E. Chrencik, H.-P. Su, Y. Gomez Llorente, R. L. Palte, D. J. Klein, R. P. Hayes, S. P. Antine, G. A. Asmar-Rovira, N. Bertoletti, N. J. Byrne, B. L. Carroll, C. D. Cordeiro, K. Curfman, J. Dutko, M. J. Eddins, T. Fischmann, B. Govindarajan, C. W. Hecksel, T. Ho, K. Hollenstein, M. R. Hong, A. Hruza, A. Ishchenko, M. Jaime-Garza, M. J. Jaremko, H. L. Knox, J. Kostas, H. Krishnamurthy, I. Manolaridis, C. L. Noland, A. T. Partridge, S. B. Patel, P. Reichert, J. C. Reid, C. Felisberto-Rodrigues, C. A. Shintre, J. M. Shipman, E. Tamilselvan, M. E. Walker, S. N. Walker, R. Wang, K. M. Wright, T.-J. Yen, B.-K. Yoo, S. B. Gabelli
Using Merck’s R&D practice as a thread, this review traces how structural biology grew from the macromolecular crystallography of the 1980s into a platform that now integrates X-ray crystallography, cryo-electron microscopy, microcrystal electron diffraction, and cryo-electron tomography. Through examples across therapeutic areas, the authors show how these complementary methods deliver structural information from atomic to cellular scale and support target discovery and validation, hit finding, lead optimization, and clinical advancement. The article also considers how structural techniques inform antibody and vaccine formulation strategies. Its focus extends further, to pairing in vitro with in situ studies: placing structure determination back into the native cellular environment could help explain drug mechanisms of action and push development of next-generation therapeutics. What the paper offers is an integrated view of multiple technologies working across the R&D pipeline.
Highlight · By integrating multiscale, complementary techniques, structural biology is extending the structural information used in drug development from isolated samples to native cellular environments.
03 · A novel fusion tool to enable G protein-coupled receptor structure determination
Acta Crystallographica Section D: Structural Biology · Published 2026-06-01
N. R. Shah, M. Oosterlaken, C. Bisson, M. Bertinelli, A. M. Churchill-Angus, A. Hutchin, J. Kopec, V. Kotov, C. R. McFarlane, E. Griss Pascualli, A. Pavic, M. Zebisch, C. Fiez-Vandal, E. Fabini, S. Duclos
To help solve the structures of G protein-coupled receptors in their inactive conformation, the researchers developed a fusion tool suited to cryo-electron microscopy. Starting from eight natural proteins drawn from the PDB, the team inserted candidates into the receptor’s third intracellular loop, preferring those with high-affinity binding partners to add marker mass. Working with the adenosine A2A receptor as a model and screening by expression, aggregation, and ligand binding, they settled on β-lactamase with its binding partner BLIPII and solved the structure in the antagonist-bound state at an overall average resolution of 3.2 Å. Local refinement sharpened details at the receptor’s orthosteric site so that ZM241385 could be modeled consistently with the published crystal structure. The tool offers another option for GPCR structure studies, and its position relative to the receptor may help reduce steric clashes; the validation case presented in the abstract is still the A2A receptor.
Highlight · A fusion marker built from β-lactamase and BLIPII delivered a 3.2 Å cryo-EM structure of the A2A receptor, offering a new tool for resolving inactive GPCRs.
04 · Autonomous biomedical research with an artificial intelligence agent
Science · Published 2026-08-20
Kexin Huang, Serena Zhang, Hanchen Wang, Yuanhao Qu, Yingzhou Lu, Ryan Li, Yusuf Roohani, Lin Qiu, Shiyi Cao, Gavin Li, Junze Zhang, Di Yin, Rick Wierenga, Deniz Kavi, Sherry Liu, Tianwei She, Shruti Marwaha, Jennefer N. Carter, Xin Zhou, Matthew T. Wheeler, Jonathan A. Bernstein, Mengdi Wang, Peng He, Jingtian Zhou, Michael P. Snyder, Le Cong, Aviv Regev, Jure Leskovec
The researchers propose Biomni, a general-purpose biomedical AI agent aimed at the repetitive, fragmented tasks in research that demand coordination across tools. From literature spanning 25 fields, the system compiles tools, databases, and experimental protocols, and combines LLM reasoning, retrieval-augmented planning, and code execution to organize workflows on the fly. In the extended summary included with this item, Biomni reached an average accuracy of 57% on 443 questions spanning ten task categories, and on three expert-level tasks matched expert accuracy in less time. Worked examples include analysis of wearable-device data, a cloning protocol that was executed experimentally and confirmed by sequencing, and code for controlling a lab robot. The study argues that a general agent can connect diverse computational and experimental resources, though these results correspond to specific benchmarks and cases and should not be read as proof that every research task can be automated reliably.
Highlight · By combining tool retrieval, dynamic planning, and code execution, Biomni links a wide range of biomedical tasks and is validated on both benchmarks and real experimental cases.
05 · Artificial intelligence in drug discovery — what it is, where we stand and the path forward
Nature Reviews Drug Discovery · Published 2026-08-07
Andreas Bender, Morgan C. Thomas, Jack W. Scannell, David A. Shaywitz, Gian Marco Ghiandoni, Joe G. Greener, Lavinia-Lorena Pruteanu, Rachel DeVay Jacobson, Koichi Handa, Mariko Hirano, Srijit Seal, Manas Mahale, Marco F. Schmidt, Tim Ahfeldt, Francesca Grisoni, Isidro Cortes-Ciriano
This perspective piece reassesses progress in AI for drug discovery, weighing it against whether AI can bring patients safer, more effective drugs faster. The authors note that even though method development and benchmarking are highly active, the evidence that AI currently produces clinically relevant impact remains limited. They analyze several underlying causes: model development pays too little attention to clinical translation, life-science data are condition-dependent, real problems are poorly defined, and computational models do not fully specify the conditions practical use demands. When technological push outruns pull from scientific questions, research directions can be skewed; turning technical capability into systems large enough and easy enough to use also takes time. The authors suggest future benchmark evaluations should probe whether AI actually improves R&D decisions, so that tools gain translational relevance and methods are assessed against the real goals of drug discovery.
Highlight · Evaluation of AI in drug discovery should move from model performance toward decision quality and translational value.
Nature Reviews Drug Discovery · Published 2026-09-01
Yee Kwan Law, Selene Glück, Enrico Girardi, Ariel Bensimon, S. Andreas Angermayr, Giulio Superti-Furga
The solute carrier family of membrane transporters already includes several established CNS drug targets, neurotransmitter reuptake inhibitors being the prime example. This review widens the lens to more SLC members: quite a few show dysregulated expression in the central nervous system and have been tied to neurological disease through genetics or functional studies. By regulating metabolite and ion flux, SLCs shape developmental programs and cellular states across areas such as epilepsy, neurodegenerative disease, and autism spectrum disorders. The authors note that progress in assigning biochemical and cellular functions to previously uncharacterized SLCs is broadening the range of targetable candidates. At the same time, emerging chemical strategies that raise or lower transporter abundance expand the options for intervention. Together these developments support SLCs as a fruitful direction for the next generation of CNS drug discovery.
Highlight · SLC target exploration is shifting from classic neurotransmitter reuptake toward a broader transport of metabolites and ions, and toward regulating transporter abundance.
07 · Effects of obesity on brain health and cognition
Nature Reviews Neurology · Published 2026-09-01
Cali M. McEntee, Masha G. Savelieff, Mohamed H. Noureldein, Stephanie A. Eid, Wolfgang Grisold, Jason J. Hassenstab, Eva L. Feldman
Obesity’s effects on brain health are drawing more attention, and among the important concerns are cognitive impairment and a higher risk of dementia. Reviewing the global epidemiology of obesity and clinical studies, this paper points out that obesity across the lifecourse is linked to cognitive decline, with midlife obesity deserving special notice. Combining preclinical and clinical evidence, the authors discuss multiple routes by which obesity may affect the brain, including hypothalamic dysregulation and metabolic, inflammatory, and vascular disturbances that act on the hippocampus and prefrontal cortex. The article also surveys obesity-related interventions that have shown cognitive benefit and, across clinical, translational, and interventional research, summarizes the evidence gaps that remain. The abstract supports bringing metabolic health into studies of cognitive risk, but it offers no effect sizes for specific interventions and cannot be used to conclude that any single weight-loss approach will reliably improve cognition.
Highlight · Central to this review are the link between midlife obesity and the risk of cognitive decline, and the joint contribution of metabolic, inflammatory, and vascular pathways.
08 · Symptomatic treatment for Alzheimer disease: current evidence and future directions
Nature Reviews Neurology · Published 2026-09-01
Clive Ballard, Pat Doherty, Zahinoor Ismail, Anne Corbett, Jeffrey L. Cummings
Even as disease-targeting therapies emerge, symptomatic treatment and management remain a central part of Alzheimer disease care. This review spans cognition, daily function, and neuropsychiatric symptoms: cognitive training, exercise, and multimodal lifestyle programs show small to moderate benefits but are hard to deliver in practice, while cholinesterase inhibitors and memantine offer limited yet meaningful gains, with biomarker stratification suggesting that people with pathologically confirmed disease may benefit more. For agitation and depression, individualized psychosocial interventions have value; atypical antipsychotics are only moderately effective and carry major safety concerns. The authors also discuss emerging drug directions, stressing that results should be interpreted alongside effect sizes and safety. Future progress rests on biomarker-guided patient selection, a sensible combination of drug and non-drug approaches, and patient-centered outcomes.
Highlight · Symptomatic treatment remains indispensable in Alzheimer disease; refining it will hinge on patient stratification, integrated interventions, and quality-of-life outcomes.
09 · AI-enhanced adaptive virtual screening of large libraries for ligand discovery
Nature Biotechnology · Published 2026-09-01
Domiziana Cecchini, AkshatKumar Nigam, Ming Tang, Joana Reis, Matt Koop, Andrea Gottinger, Callum Robert Nicoll, Yao Wang, Abhilash Jayaraj, Süleyman Selim Çinarogluˋ, Ricarda Törner, Yehor Malets, Minko Gehev, Krishna M. Padmanabha Das, Kelly Churion, Jongwan Kim, Nidhin Thomas, Yong Li, Hyuk-Soo Seo, Sirano Dhe-Paganon, Christopher Secker, Mohammad Haddadnia, Alexander Hasson, Minkai Li, Abhishek Kumar, Roni Levin-Konigsberg, Eun-Bee Choi, Geoffrey I. Shapiro, Huel Cox, Luke Sebastian, Chelsea Braithwaite, Puspalata Bashyal, Dmytro S. Radchenko, Aditya Kumar, Lei Yang, Pierre-Yves Aquilanti, Henry Gabb, Amr Alhossary, Eric O’Neill, Gerhard Wagner, Alán Aspuru-Guzik, Yurii S. Moroz, Charalampos G. Kalodimos, Konstantin Fackeldey, John D. Schuetz, Andrea Mattevi, Haribabu Arthanari, Christoph Gorgulla
Ultra-large-scale virtual screening can assess billions of molecules, but cost, flexibility, and scalability remain hurdles. The researchers propose AdaptiveFlow, an open-source platform, bundled with a docking-ready Enamine REAL Space set containing 69 billion compounds. The platform uses an 18-dimensional grid of molecular properties to prioritize promising chemical subspaces, can incorporate active learning, and cuts computational cost by orders of magnitude. The system integrates more than 1,500 docking configurations, including GPU-accelerated and machine-learning approaches, and is reported to scale near-linearly to 5.6 million CPUs on AWS. The team obtained nanomolar inhibitors against FSP1 and PARP1 and dissected the FSP1 inhibition mechanism through co-crystal structures. The work delivers both large-scale screening infrastructure and instances of experimental validation, offering a platform on which future AI methods can be developed.
Highlight · AdaptiveFlow couples adaptive screening of chemical space with large-scale computing, and its value is validated by nanomolar inhibitors of FSP1 and PARP1.
This issue rounds up nine papers added to my personal Zotero library between 31 August and 6 September 2026, listed in reverse order of accession. All dates use Beijing time and refer to when an item was added; each paper’s publication date is given separately.
The notes below were assembled by AI from the abstracts; the Biomni entry includes an extended summary. Full texts were not read. Overviews and Highlights are meant for quick scanning — please consult the original papers for methods and findings.
01 · Phage therapy: from basic biology to clinical application
Nature Reviews Microbiology · Published 2026-08-27
Holly A. Sinclair, Ruby C. Y. Lin, Nouri L. Ben Zakour, Paul L. Bollyky, Jonathan R. Iredell
Bacteriophages have been used to treat bacterial infections in humans for over a century, yet in vitro results still predict clinical outcomes poorly and no consensus has formed on how best to use them. This review starts from phage biology to discuss how to choose phages rationally and understand the interplay among bacteria, phages, and the human host. The authors stress that the adaptations bacteria and phages undergo in vivo must be built into treatment design, and that paradigms borrowed from conventional antibiotics cannot simply be copied over. They set out clinical strategies for different infections with their strengths and limits, and also discuss obstacles in manufacturing, scaling, and regulation. How phages interact with the human immune system is another key determinant of whether treatment succeeds. The review casts phages as an important avenue against antimicrobial resistance, but one whose use depends on deeper mechanistic understanding and case-specific choices.
Highlight · Phage therapy has to reckon with the dynamic relationship between bacteria, phages, and the host; in vitro activity cannot be equated with clinical efficacy.
DOI: 10.1038/s41579-026-01352-5
02 · The evolving role of structural biology in pharma: integration of X-ray crystallography, cryo-electron microscopy and beyond
Acta Crystallographica Section D: Structural Biology · Published 2026-06-01
J. E. Chrencik, H.-P. Su, Y. Gomez Llorente, R. L. Palte, D. J. Klein, R. P. Hayes, S. P. Antine, G. A. Asmar-Rovira, N. Bertoletti, N. J. Byrne, B. L. Carroll, C. D. Cordeiro, K. Curfman, J. Dutko, M. J. Eddins, T. Fischmann, B. Govindarajan, C. W. Hecksel, T. Ho, K. Hollenstein, M. R. Hong, A. Hruza, A. Ishchenko, M. Jaime-Garza, M. J. Jaremko, H. L. Knox, J. Kostas, H. Krishnamurthy, I. Manolaridis, C. L. Noland, A. T. Partridge, S. B. Patel, P. Reichert, J. C. Reid, C. Felisberto-Rodrigues, C. A. Shintre, J. M. Shipman, E. Tamilselvan, M. E. Walker, S. N. Walker, R. Wang, K. M. Wright, T.-J. Yen, B.-K. Yoo, S. B. Gabelli
Using Merck’s R&D practice as a thread, this review traces how structural biology grew from the macromolecular crystallography of the 1980s into a platform that now integrates X-ray crystallography, cryo-electron microscopy, microcrystal electron diffraction, and cryo-electron tomography. Through examples across therapeutic areas, the authors show how these complementary methods deliver structural information from atomic to cellular scale and support target discovery and validation, hit finding, lead optimization, and clinical advancement. The article also considers how structural techniques inform antibody and vaccine formulation strategies. Its focus extends further, to pairing in vitro with in situ studies: placing structure determination back into the native cellular environment could help explain drug mechanisms of action and push development of next-generation therapeutics. What the paper offers is an integrated view of multiple technologies working across the R&D pipeline.
Highlight · By integrating multiscale, complementary techniques, structural biology is extending the structural information used in drug development from isolated samples to native cellular environments.
DOI: 10.1107/S2059798326004390
03 · A novel fusion tool to enable G protein-coupled receptor structure determination
Acta Crystallographica Section D: Structural Biology · Published 2026-06-01
N. R. Shah, M. Oosterlaken, C. Bisson, M. Bertinelli, A. M. Churchill-Angus, A. Hutchin, J. Kopec, V. Kotov, C. R. McFarlane, E. Griss Pascualli, A. Pavic, M. Zebisch, C. Fiez-Vandal, E. Fabini, S. Duclos
To help solve the structures of G protein-coupled receptors in their inactive conformation, the researchers developed a fusion tool suited to cryo-electron microscopy. Starting from eight natural proteins drawn from the PDB, the team inserted candidates into the receptor’s third intracellular loop, preferring those with high-affinity binding partners to add marker mass. Working with the adenosine A2A receptor as a model and screening by expression, aggregation, and ligand binding, they settled on β-lactamase with its binding partner BLIPII and solved the structure in the antagonist-bound state at an overall average resolution of 3.2 Å. Local refinement sharpened details at the receptor’s orthosteric site so that ZM241385 could be modeled consistently with the published crystal structure. The tool offers another option for GPCR structure studies, and its position relative to the receptor may help reduce steric clashes; the validation case presented in the abstract is still the A2A receptor.
Highlight · A fusion marker built from β-lactamase and BLIPII delivered a 3.2 Å cryo-EM structure of the A2A receptor, offering a new tool for resolving inactive GPCRs.
DOI: 10.1107/S2059798326003785
04 · Autonomous biomedical research with an artificial intelligence agent
Science · Published 2026-08-20
Kexin Huang, Serena Zhang, Hanchen Wang, Yuanhao Qu, Yingzhou Lu, Ryan Li, Yusuf Roohani, Lin Qiu, Shiyi Cao, Gavin Li, Junze Zhang, Di Yin, Rick Wierenga, Deniz Kavi, Sherry Liu, Tianwei She, Shruti Marwaha, Jennefer N. Carter, Xin Zhou, Matthew T. Wheeler, Jonathan A. Bernstein, Mengdi Wang, Peng He, Jingtian Zhou, Michael P. Snyder, Le Cong, Aviv Regev, Jure Leskovec
The researchers propose Biomni, a general-purpose biomedical AI agent aimed at the repetitive, fragmented tasks in research that demand coordination across tools. From literature spanning 25 fields, the system compiles tools, databases, and experimental protocols, and combines LLM reasoning, retrieval-augmented planning, and code execution to organize workflows on the fly. In the extended summary included with this item, Biomni reached an average accuracy of 57% on 443 questions spanning ten task categories, and on three expert-level tasks matched expert accuracy in less time. Worked examples include analysis of wearable-device data, a cloning protocol that was executed experimentally and confirmed by sequencing, and code for controlling a lab robot. The study argues that a general agent can connect diverse computational and experimental resources, though these results correspond to specific benchmarks and cases and should not be read as proof that every research task can be automated reliably.
Highlight · By combining tool retrieval, dynamic planning, and code execution, Biomni links a wide range of biomedical tasks and is validated on both benchmarks and real experimental cases.
DOI: 10.1126/science.adz4351
05 · Artificial intelligence in drug discovery — what it is, where we stand and the path forward
Nature Reviews Drug Discovery · Published 2026-08-07
Andreas Bender, Morgan C. Thomas, Jack W. Scannell, David A. Shaywitz, Gian Marco Ghiandoni, Joe G. Greener, Lavinia-Lorena Pruteanu, Rachel DeVay Jacobson, Koichi Handa, Mariko Hirano, Srijit Seal, Manas Mahale, Marco F. Schmidt, Tim Ahfeldt, Francesca Grisoni, Isidro Cortes-Ciriano
This perspective piece reassesses progress in AI for drug discovery, weighing it against whether AI can bring patients safer, more effective drugs faster. The authors note that even though method development and benchmarking are highly active, the evidence that AI currently produces clinically relevant impact remains limited. They analyze several underlying causes: model development pays too little attention to clinical translation, life-science data are condition-dependent, real problems are poorly defined, and computational models do not fully specify the conditions practical use demands. When technological push outruns pull from scientific questions, research directions can be skewed; turning technical capability into systems large enough and easy enough to use also takes time. The authors suggest future benchmark evaluations should probe whether AI actually improves R&D decisions, so that tools gain translational relevance and methods are assessed against the real goals of drug discovery.
Highlight · Evaluation of AI in drug discovery should move from model performance toward decision quality and translational value.
DOI: 10.1038/s41573-026-01496-2
06 · Solute carrier membrane transporters: emerging targets in CNS disorders
Nature Reviews Drug Discovery · Published 2026-09-01
Yee Kwan Law, Selene Glück, Enrico Girardi, Ariel Bensimon, S. Andreas Angermayr, Giulio Superti-Furga
The solute carrier family of membrane transporters already includes several established CNS drug targets, neurotransmitter reuptake inhibitors being the prime example. This review widens the lens to more SLC members: quite a few show dysregulated expression in the central nervous system and have been tied to neurological disease through genetics or functional studies. By regulating metabolite and ion flux, SLCs shape developmental programs and cellular states across areas such as epilepsy, neurodegenerative disease, and autism spectrum disorders. The authors note that progress in assigning biochemical and cellular functions to previously uncharacterized SLCs is broadening the range of targetable candidates. At the same time, emerging chemical strategies that raise or lower transporter abundance expand the options for intervention. Together these developments support SLCs as a fruitful direction for the next generation of CNS drug discovery.
Highlight · SLC target exploration is shifting from classic neurotransmitter reuptake toward a broader transport of metabolites and ions, and toward regulating transporter abundance.
DOI: 10.1038/s41573-026-01513-4
07 · Effects of obesity on brain health and cognition
Nature Reviews Neurology · Published 2026-09-01
Cali M. McEntee, Masha G. Savelieff, Mohamed H. Noureldein, Stephanie A. Eid, Wolfgang Grisold, Jason J. Hassenstab, Eva L. Feldman
Obesity’s effects on brain health are drawing more attention, and among the important concerns are cognitive impairment and a higher risk of dementia. Reviewing the global epidemiology of obesity and clinical studies, this paper points out that obesity across the lifecourse is linked to cognitive decline, with midlife obesity deserving special notice. Combining preclinical and clinical evidence, the authors discuss multiple routes by which obesity may affect the brain, including hypothalamic dysregulation and metabolic, inflammatory, and vascular disturbances that act on the hippocampus and prefrontal cortex. The article also surveys obesity-related interventions that have shown cognitive benefit and, across clinical, translational, and interventional research, summarizes the evidence gaps that remain. The abstract supports bringing metabolic health into studies of cognitive risk, but it offers no effect sizes for specific interventions and cannot be used to conclude that any single weight-loss approach will reliably improve cognition.
Highlight · Central to this review are the link between midlife obesity and the risk of cognitive decline, and the joint contribution of metabolic, inflammatory, and vascular pathways.
DOI: 10.1038/s41582-026-01251-6
08 · Symptomatic treatment for Alzheimer disease: current evidence and future directions
Nature Reviews Neurology · Published 2026-09-01
Clive Ballard, Pat Doherty, Zahinoor Ismail, Anne Corbett, Jeffrey L. Cummings
Even as disease-targeting therapies emerge, symptomatic treatment and management remain a central part of Alzheimer disease care. This review spans cognition, daily function, and neuropsychiatric symptoms: cognitive training, exercise, and multimodal lifestyle programs show small to moderate benefits but are hard to deliver in practice, while cholinesterase inhibitors and memantine offer limited yet meaningful gains, with biomarker stratification suggesting that people with pathologically confirmed disease may benefit more. For agitation and depression, individualized psychosocial interventions have value; atypical antipsychotics are only moderately effective and carry major safety concerns. The authors also discuss emerging drug directions, stressing that results should be interpreted alongside effect sizes and safety. Future progress rests on biomarker-guided patient selection, a sensible combination of drug and non-drug approaches, and patient-centered outcomes.
Highlight · Symptomatic treatment remains indispensable in Alzheimer disease; refining it will hinge on patient stratification, integrated interventions, and quality-of-life outcomes.
DOI: 10.1038/s41582-026-01260-5
09 · AI-enhanced adaptive virtual screening of large libraries for ligand discovery
Nature Biotechnology · Published 2026-09-01
Domiziana Cecchini, AkshatKumar Nigam, Ming Tang, Joana Reis, Matt Koop, Andrea Gottinger, Callum Robert Nicoll, Yao Wang, Abhilash Jayaraj, Süleyman Selim Çinarogluˋ, Ricarda Törner, Yehor Malets, Minko Gehev, Krishna M. Padmanabha Das, Kelly Churion, Jongwan Kim, Nidhin Thomas, Yong Li, Hyuk-Soo Seo, Sirano Dhe-Paganon, Christopher Secker, Mohammad Haddadnia, Alexander Hasson, Minkai Li, Abhishek Kumar, Roni Levin-Konigsberg, Eun-Bee Choi, Geoffrey I. Shapiro, Huel Cox, Luke Sebastian, Chelsea Braithwaite, Puspalata Bashyal, Dmytro S. Radchenko, Aditya Kumar, Lei Yang, Pierre-Yves Aquilanti, Henry Gabb, Amr Alhossary, Eric O’Neill, Gerhard Wagner, Alán Aspuru-Guzik, Yurii S. Moroz, Charalampos G. Kalodimos, Konstantin Fackeldey, John D. Schuetz, Andrea Mattevi, Haribabu Arthanari, Christoph Gorgulla
Ultra-large-scale virtual screening can assess billions of molecules, but cost, flexibility, and scalability remain hurdles. The researchers propose AdaptiveFlow, an open-source platform, bundled with a docking-ready Enamine REAL Space set containing 69 billion compounds. The platform uses an 18-dimensional grid of molecular properties to prioritize promising chemical subspaces, can incorporate active learning, and cuts computational cost by orders of magnitude. The system integrates more than 1,500 docking configurations, including GPU-accelerated and machine-learning approaches, and is reported to scale near-linearly to 5.6 million CPUs on AWS. The team obtained nanomolar inhibitors against FSP1 and PARP1 and dissected the FSP1 inhibition mechanism through co-crystal structures. The work delivers both large-scale screening infrastructure and instances of experimental validation, offering a platform on which future AI methods can be developed.
Highlight · AdaptiveFlow couples adaptive screening of chemical space with large-scale computing, and its value is validated by nanomolar inhibitors of FSP1 and PARP1.
DOI: 10.1038/s41587-026-03217-x