[{"data":1,"prerenderedAt":316},["ShallowReactive",2],{"navigation":3,"author benedict-w-j-irwin":203,"allBlogPosts benedict-w-j-irwin":299},{"data":4},{"consentBanner":5,"contact":10,"setting":131},{"title":6,"approveButtonText":7,"rejectButtonText":8,"description":9},"Privacy preferences","Accept","Reject","We and selected third parties use cookies or similar technologies for technical purposes and, with your consent, for “measurement” and “targeting & advertising” as specified in the [privacy & cookie policy.](\u002Fprivacy-policy) Denying consent may make related features unavailable.\n\nYou can freely give, deny, or withdraw your consent at any time.\n\nUse the “Accept” button to consent to the use of such technologies. Use the “Reject” button to continue without accepting.",{"title":11,"body":12},"Get in touch",{"value":13,"blocks":27},{"schema":14,"document":15},"dast",{"type":16,"children":17},"root",[18,21],{"item":19,"type":20},"XUHV-8w0S--Q4LJ4iFBxcQ","block",{"type":22,"children":23},"paragraph",[24],{"type":25,"value":26},"span","",[28],{"__typename":29,"id":19,"top":30,"middleAlignment":31,"middleClasses":26,"middleGrid":32,"bottom":30},"SectionBlockRecord",null,"center",[33,87],{"body":34},{"value":35,"blocks":63},{"schema":14,"document":36},{"type":16,"children":37},[38,43,45,49,59],{"type":22,"children":39},[40],{"item":41,"type":42},"aUH-zdl4SG-L-DG5FECu6w","inlineBlock",{"item":44,"type":20},"Kc-4dNK3Ry2-lC6eeQasIg",{"type":22,"children":46},[47],{"type":25,"value":48},"Improve AI model performance with life sciences data networks by securely connecting proprietary life sciences data while preserving confidentiality and IP",{"type":22,"children":50},[51,53,57],{"type":25,"value":52},"General enquiries: ",{"type":25,"marks":54,"value":56},[55],"emphasis","info@apheris.com",{"type":25,"value":58}," Media: press@apheris.com",{"type":22,"children":60},[61],{"type":25,"value":62},"Although most of us work remotely, we’re headquartered in Berlin. You can find us at:\nc\u002Fo Mindspace, Skalitzer Str. 104, 10997 Berlin, Germany",[64,77],{"__typename":65,"id":44,"numberOfColumns":66,"centered":30,"fontSize":67,"body":68},"BlockParagraphRecord","1","2xl",{"value":69,"blocks":76},{"schema":14,"document":70},{"type":16,"children":71},[72],{"type":22,"children":73},[74],{"type":25,"value":75},"Join or build a data network",[],{"__typename":78,"id":41,"size":79,"link":26,"controls":80,"autoPlay":80,"image":81},"BlockInlineImageRecord","grow and shrink",false,{"responsiveImage":30,"height":82,"width":83,"format":84,"url":85,"alt":86},175,456,"svg","https:\u002F\u002Fwww.datocms-assets.com\u002F77111\u002F1694086017-contact-cta.svg","Contact us at Apheris",{"body":88},{"value":89,"blocks":94},{"schema":14,"document":90},{"type":16,"children":91},[92],{"item":93,"type":20},"dKSU5XqcTqeDO1ic6iqg0w",[95],{"__typename":96,"id":93,"hubspotFormId":97,"hubspotPortalId":98,"formFieldsList":99},"HubspotFormRecord","dd8142e3-875c-4b4e-b303-a02d621f0e24","8944365",[100,107,112,117,122,127],{"hubspotFieldName":101,"label":102,"fieldType":103,"rules":104},"firstname","First name","text",[105],{"formRule":106},"required",{"hubspotFieldName":108,"label":109,"fieldType":103,"rules":110},"lastname","Last name",[111],{"formRule":106},{"hubspotFieldName":113,"label":114,"fieldType":103,"rules":115},"email","Email",[116],{"formRule":113},{"hubspotFieldName":118,"label":119,"fieldType":103,"rules":120},"company","Company",[121],{"formRule":106},{"hubspotFieldName":123,"label":124,"fieldType":125,"rules":126},"message","Message","textarea",[],{"hubspotFieldName":128,"label":129,"fieldType":125,"rules":130},"self_attribution_message","How did you hear about us? (e.g., Google, LinkedIn, etc.)",[],{"navigation":132},{"navigation":133,"menus":144},[134,137,140,143],{"label":135,"id":136},"Product","product",{"label":138,"id":139},"Networks","join",{"label":141,"id":142},"Applications","applications",{"label":119,"id":118},{"product":145,"applications":163,"join":168,"company":188},[146,153,158],{"icon":147,"label":148,"subtitle":149,"hideArrow":150,"centerText":150,"largeIcon":150,"largeText":150,"classesMobile":151,"link":152},"i-app:shield","Apheris Gateway","Federated computing",true,"row-span-2","\u002Fproduct",{"icon":154,"label":155,"subtitle":156,"hideArrow":150,"largeIcon":150,"link":157},"i-app:book","Documentation","Drug discovery AI applications powered by federated networks","\u002Fdocs",{"icon":159,"label":160,"subtitle":161,"hideArrow":150,"largeIcon":150,"link":162},"i-app:target","Trust Center","One-stop shop for compliance docs and audit reports.","https:\u002F\u002Ftrust.apheris.com\u002F",[164],{"label":165,"subtitle":166,"link":167},"ApherisFold","Co-Folding","\u002Fapplications\u002Fapherisfold",[169,174,178,183],{"icon":170,"label":171,"labelMobile":172,"link":173},"i-app:ai","AI Structural Biology Network","AISB Network","\u002Fjoin-a-network\u002Faisb",{"icon":175,"label":176,"link":177},"i-app:pharma","ADMET Network","\u002Fjoin-a-network\u002Fadmet",{"icon":179,"label":180,"labelMobile":181,"link":182},"i-app:antibody","Antibody Developability Network","Antibody","\u002Fjoin-a-network\u002Fantibody-developability-consortium",{"icon":184,"label":185,"labelMobile":186,"link":187},"i-app:collaborate","RWE Network","RWE","\u002Fjoin-a-network\u002Ffederated-real-world-evidence-rwe-network",[189,195,200],{"label":190,"subtitle":191,"link":192,"hideArrow":150,"largeText":150,"classes":193,"classesMobile":194},"Resources","Discover news, articles, white papers, and other materials related to the company activities.","\u002Fresources","col-span-3 row-span-3","row-span-3",{"icon":196,"label":197,"link":198,"classes":199},"i-app:marketplace","About","\u002Fcompany","col-span-2",{"icon":184,"label":201,"link":202,"classes":199},"Careers","\u002Fcareers",{"data":204},{"postPage":205,"authorPage":221,"page":223},{"ctaNewsletter":206},[207],{"id":208,"title":209,"image":210,"mobileImage":217},"55418820","Insights delivered to your inbox monthly",{"responsiveImage":211,"format":215,"url":216,"alt":214},{"srcSet":212,"webpSrcSet":213,"alt":214},"https:\u002F\u002Fwww.datocms-assets.com\u002F77111\u002F1745520894-molecule-connected-data.png?dpr=0.25&fit=fill&max-h=250&max-w=250 256w,https:\u002F\u002Fwww.datocms-assets.com\u002F77111\u002F1745520894-molecule-connected-data.png?dpr=0.5&fit=fill&max-h=500&max-w=500 512w,https:\u002F\u002Fwww.datocms-assets.com\u002F77111\u002F1745520894-molecule-connected-data.png?dpr=0.75&fit=fill&max-h=750&max-w=750 768w,https:\u002F\u002Fwww.datocms-assets.com\u002F77111\u002F1745520894-molecule-connected-data.png?fit=fill&max-h=1000&max-w=1000 1024w","https:\u002F\u002Fwww.datocms-assets.com\u002F77111\u002F1745520894-molecule-connected-data.png?dpr=0.25&fit=fill&fm=webp&max-h=250&max-w=250 256w,https:\u002F\u002Fwww.datocms-assets.com\u002F77111\u002F1745520894-molecule-connected-data.png?dpr=0.5&fit=fill&fm=webp&max-h=500&max-w=500 512w,https:\u002F\u002Fwww.datocms-assets.com\u002F77111\u002F1745520894-molecule-connected-data.png?dpr=0.75&fit=fill&fm=webp&max-h=750&max-w=750 768w,https:\u002F\u002Fwww.datocms-assets.com\u002F77111\u002F1745520894-molecule-connected-data.png?fit=fill&fm=webp&max-h=1000&max-w=1000 1024w","Molecule connected data Apheris","png","https:\u002F\u002Fwww.datocms-assets.com\u002F77111\u002F1745520894-molecule-connected-data.png",{"responsiveImage":218,"url":216},{"srcSet":219,"webpSrcSet":220},"https:\u002F\u002Fwww.datocms-assets.com\u002F77111\u002F1745520894-molecule-connected-data.png?dpr=0.25&fit=fill&max-h=200&max-w=200 256w,https:\u002F\u002Fwww.datocms-assets.com\u002F77111\u002F1745520894-molecule-connected-data.png?dpr=0.5&fit=fill&max-h=400&max-w=400 512w,https:\u002F\u002Fwww.datocms-assets.com\u002F77111\u002F1745520894-molecule-connected-data.png?dpr=0.75&fit=fill&max-h=600&max-w=600 768w,https:\u002F\u002Fwww.datocms-assets.com\u002F77111\u002F1745520894-molecule-connected-data.png?fit=fill&max-h=800&max-w=800 1024w","https:\u002F\u002Fwww.datocms-assets.com\u002F77111\u002F1745520894-molecule-connected-data.png?dpr=0.25&fit=fill&fm=webp&max-h=200&max-w=200 256w,https:\u002F\u002Fwww.datocms-assets.com\u002F77111\u002F1745520894-molecule-connected-data.png?dpr=0.5&fit=fill&fm=webp&max-h=400&max-w=400 512w,https:\u002F\u002Fwww.datocms-assets.com\u002F77111\u002F1745520894-molecule-connected-data.png?dpr=0.75&fit=fill&fm=webp&max-h=600&max-w=600 768w,https:\u002F\u002Fwww.datocms-assets.com\u002F77111\u002F1745520894-molecule-connected-data.png?fit=fill&fm=webp&max-h=800&max-w=800 1024w",{"titleArticles":222},"All Articles by ",{"seo":224,"id":289,"fullName":226,"jobTitle":290,"description":291,"linkedIn":292,"image":293},[225,228,232,235,239,242,245,249,253,256,260,263,266,270,274,278,282,285],{"attributes":30,"content":226,"tag":227},"Benedict W. J. Irwin","title",{"attributes":229,"content":30,"tag":231},{"property":230,"content":226},"og:title","meta",{"attributes":233,"content":30,"tag":231},{"name":234,"content":226},"twitter:title",{"attributes":236,"content":30,"tag":231},{"name":237,"content":238},"description","From federated models to drug discovery decisions. We deliver drug discovery AI models through secure, local applications.",{"attributes":240,"content":30,"tag":231},{"property":241,"content":238},"og:description",{"attributes":243,"content":30,"tag":231},{"name":244,"content":238},"twitter:description",{"attributes":246,"content":30,"tag":231},{"property":247,"content":248},"og:image","https:\u002F\u002Fwww.datocms-assets.com\u002F77111\u002F1760344330-ben-irwin-apheris-team.jpg?auto=format&fit=max&w=1200",{"attributes":250,"content":30,"tag":231},{"property":251,"content":252},"og:image:width","500",{"attributes":254,"content":30,"tag":231},{"property":255,"content":252},"og:image:height",{"attributes":257,"content":30,"tag":231},{"property":258,"content":259},"og:image:alt","Ben Irwin Apheris team",{"attributes":261,"content":30,"tag":231},{"name":262,"content":248},"twitter:image",{"attributes":264,"content":30,"tag":231},{"name":265,"content":259},"twitter:image:alt",{"attributes":267,"content":30,"tag":231},{"property":268,"content":269},"og:locale","en",{"attributes":271,"content":30,"tag":231},{"property":272,"content":273},"og:type","article",{"attributes":275,"content":30,"tag":231},{"property":276,"content":277},"og:site_name","Apheris",{"attributes":279,"content":30,"tag":231},{"property":280,"content":281},"article:modified_time","2025-09-19T13:17:35Z",{"attributes":283,"content":30,"tag":231},{"property":284,"content":26},"article:publisher",{"attributes":286,"content":30,"tag":231},{"name":287,"content":288},"twitter:card","summary","MHZLLKAqQh6x_5Evm_GfLg","Principal ML Engineer","AI\u002FML and simulations based solutions for drug discovery. \n\nI am familiar with a range of conventional ML methods, as well as deep learning, sparse data methods, large datasets, imputation, graph convolution networks, transformers, recurrent networks, generative methods, recommender systems, kernel density methods and distribution matching, Bayesian optimisation, Gaussian processes, and generalised additive models. I have some knowledge of time series modelling, signal processing methods, image based methods. I am mathematically fluent and can comfortably generate new solutions where there is little to no existing literature. This extends to non-machine learning algorithms such as integer programming, trees and data structures.\n\nI completed my PhD in the Theory of Condensed Matter (TCM) group at the Cavendish Laboratory at the University of Cambridge. I derived the Atomwise Free Energy Perturbation (AFEP) method to calculate approximate decompositions of the free energy of solvation and binding and have run thousands of molecular dynamics simulations. I am familiar with free energy perturbation, thermodynamic integration various computational chemistry and simulation methods.\n\nI am familiar with AWS, AzureML\u002FADO, high performance computing, parallel algorithms (OpenMP, MPI) and software development in a variety of languages (C, C++, Python, Perl). I have strong mathematical skills and can use various computer algebra systems (Mathematica, Maple, Matlab). I'm generally interested in mathematics, machine learning and artificial intelligence.\n\nI have reviewed journal papers for JMLR, JCIM, In Silico Pharmacology, and have made many additions to the OEIS.AI\u002FML and simulations based solutions for drug discovery. I am familiar with a range of conventional ML methods, as well as deep learning, sparse data methods, large datasets, imputation, graph convolution networks, transformers, recurrent networks, generative methods, recommender systems, kernel density methods and distribution matching, Bayesian optimisation, Gaussian processes, and generalised additive models. I have some knowledge of time series modelling, signal processing methods, image based methods. I am mathematically fluent and can comfortably generate new solutions where there is little to no existing literature. This extends to non-machine learning algorithms such as integer programming, trees and data structures. I completed my PhD in the Theory of Condensed Matter (TCM) group at the Cavendish Laboratory at the University of Cambridge. I derived the Atomwise Free Energy Perturbation (AFEP) method to calculate approximate decompositions of the free energy of solvation and binding and have run thousands of molecular dynamics simulations. I am familiar with free energy perturbation, thermodynamic integration various computational chemistry and simulation methods. I am familiar with AWS, AzureML\u002FADO, high performance computing, parallel algorithms (OpenMP, MPI) and software development in a variety of languages (C, C++, Python, Perl). I have strong mathematical skills and can use various computer algebra systems (Mathematica, Maple, Matlab). I'm generally interested in mathematics, machine learning and artificial intelligence. I have reviewed journal papers for JMLR, JCIM, In Silico Pharmacology, and have made many additions to the OEIS.","https:\u002F\u002Fwww.linkedin.com\u002Fin\u002Fbenedict-w-j-irwin-%E5%A5%94%E5%B0%8F%E5%BA%B7-5b85b124\u002F",{"responsiveImage":294,"format":297,"url":298,"alt":259},{"srcSet":295,"webpSrcSet":296,"alt":259},"https:\u002F\u002Fwww.datocms-assets.com\u002F77111\u002F1760344330-ben-irwin-apheris-team.jpg?dpr=0.25&fit=fill&max-h=250&max-w=250 125w,https:\u002F\u002Fwww.datocms-assets.com\u002F77111\u002F1760344330-ben-irwin-apheris-team.jpg?dpr=0.5&fit=fill&max-h=500&max-w=500 250w,https:\u002F\u002Fwww.datocms-assets.com\u002F77111\u002F1760344330-ben-irwin-apheris-team.jpg?dpr=0.75&fit=fill&max-h=750&max-w=750 375w,https:\u002F\u002Fwww.datocms-assets.com\u002F77111\u002F1760344330-ben-irwin-apheris-team.jpg?fit=fill&max-h=1000&max-w=1000 500w","https:\u002F\u002Fwww.datocms-assets.com\u002F77111\u002F1760344330-ben-irwin-apheris-team.jpg?dpr=0.25&fit=fill&fm=webp&max-h=250&max-w=250 125w,https:\u002F\u002Fwww.datocms-assets.com\u002F77111\u002F1760344330-ben-irwin-apheris-team.jpg?dpr=0.5&fit=fill&fm=webp&max-h=500&max-w=500 250w,https:\u002F\u002Fwww.datocms-assets.com\u002F77111\u002F1760344330-ben-irwin-apheris-team.jpg?dpr=0.75&fit=fill&fm=webp&max-h=750&max-w=750 375w,https:\u002F\u002Fwww.datocms-assets.com\u002F77111\u002F1760344330-ben-irwin-apheris-team.jpg?fit=fill&fm=webp&max-h=1000&max-w=1000 500w","jpg","https:\u002F\u002Fwww.datocms-assets.com\u002F77111\u002F1760344330-ben-irwin-apheris-team.jpg",{"data":300},{"allBlogPosts":301},[302],{"__typename":303,"slug":304,"title":305,"description":306,"contentType":307,"image":310},"BlogPostRecord","fine-tuning-the-openfold3-affinity-head-on-a-small-jak2-macrocycle-dataset","Fine-tuning the OpenFold3 affinity head on a small JAK2 macrocycle dataset","We rebuilt the SandboxAQ affinity head inside ApherisFold and fine-tuned it on 49 JAK2 macrocycles. Architectural changes alone lifted Spearman ranking from 0.418 to 0.60; fine-tuning then pushed validation Pearson from 0.23 to 0.76, mostly by correcting how the model handled inactives.",[308],{"contentType":309},"Article",{"responsiveImage":311,"format":215,"url":315,"alt":314},{"srcSet":312,"webpSrcSet":313,"alt":314},"https:\u002F\u002Fwww.datocms-assets.com\u002F77111\u002F1765807740-sik-3-ampk-selectivity-study.png?dpr=0.25&fit=fill&max-h=250&max-w=250 256w,https:\u002F\u002Fwww.datocms-assets.com\u002F77111\u002F1765807740-sik-3-ampk-selectivity-study.png?dpr=0.5&fit=fill&max-h=500&max-w=500 512w,https:\u002F\u002Fwww.datocms-assets.com\u002F77111\u002F1765807740-sik-3-ampk-selectivity-study.png?dpr=0.75&fit=fill&max-h=750&max-w=750 768w,https:\u002F\u002Fwww.datocms-assets.com\u002F77111\u002F1765807740-sik-3-ampk-selectivity-study.png?fit=fill&max-h=1000&max-w=1000 1024w","https:\u002F\u002Fwww.datocms-assets.com\u002F77111\u002F1765807740-sik-3-ampk-selectivity-study.png?dpr=0.25&fit=fill&fm=webp&max-h=250&max-w=250 256w,https:\u002F\u002Fwww.datocms-assets.com\u002F77111\u002F1765807740-sik-3-ampk-selectivity-study.png?dpr=0.5&fit=fill&fm=webp&max-h=500&max-w=500 512w,https:\u002F\u002Fwww.datocms-assets.com\u002F77111\u002F1765807740-sik-3-ampk-selectivity-study.png?dpr=0.75&fit=fill&fm=webp&max-h=750&max-w=750 768w,https:\u002F\u002Fwww.datocms-assets.com\u002F77111\u002F1765807740-sik-3-ampk-selectivity-study.png?fit=fill&fm=webp&max-h=1000&max-w=1000 1024w","SIK-3 - AMPK Selectivity Study","https:\u002F\u002Fwww.datocms-assets.com\u002F77111\u002F1765807740-sik-3-ampk-selectivity-study.png",1779266354631]