Machine-learned interatomic potentials (MLIPs), or simply machine learning potentials (MLPs), are interatomic potentials constructed using machine learning. Beginning in the 1990s, researchers have employed such programs to construct interatomic potentials by mapping atomic structures to their potential energies. These potentials are referred to as MLIPs or MLPs. Such machine learning potentials promised to fill the gap between density functional theory, a highly accurate but computationally intensive modelling method, and empirically derived or intuitively-approximated potentials, which were far lighter computationally but substantially less accurate. Improvements in artificial intelligence technology heightened the accuracy of MLPs while lowering their computational cost, increasing the role of machine learning in fitting potentials. Machine learning potentials began by using neural networks to tackle low-dimensional systems. While promising, these models could not systematically account for interatomic energy interactions; they could be applied to small molecules in a vacuum, or molecules interacting with frozen surfaces, but not much else – and even in these applications, the models often relied on force fields or potentials derived empirically or with simulations. These models thus remained confined to academia. Modern neural networks construct highly accurate and computationally light potentials, as theoretical understanding of materials science was increasingly built into their architectures and preprocessing. Almost all are local, accounting for all interactions between an atom and its neighbor up to some cutoff radius. There exist some nonlocal models, but these have been experimental for almost a decade. For most systems, reasonable cutoff radii enable highly accurate results. Almost all neural networks intake atomic coordinates and output potential energies. For some, these atomic coordinates are converted into atom-centered symmetry functions. From this data, a separate atomic neural network is trained for each element; each atomic network is evaluated whenever that element occurs in the given structure, and then the results are pooled together at the end. This process – in particular, the atom-centered symmetry functions which convey translational, rotational, and permutational invariances – has greatly improved machine learning potentials by significantly constraining the neural network search space. Other models use a similar process but emphasize bonds over atoms, using pair symmetry functions and training one network per atom pair. Other models to learn their own descriptors rather than using predetermined symmetry-dictating functions. These models, called message-passing neural networks (MPNNs), are graph neural networks. Treating molecules as three-dimensional graphs (where atoms are nodes and bonds are edges), the model takes feature vectors describing the atoms as input, and iteratively updates these vectors as information about neighboring atoms is processed through message functions and convolutions. These feature vectors are then used to predict the final potentials. The flexibility of this method often results in stronger, more generalizable models. In 2017, the first-ever MPNN model (a deep tensor neural network) was used to calculate the properties of small organic molecules. == Gaussian Approximation Potential (GAP) == One popular class of machine-learned interatomic potential is the Gaussian Approximation Potential (GAP), which combines compact descriptors of local atomic environments with Gaussian process regression to machine learn the potential energy surface of a given system. To date, the GAP framework has been used to successfully develop a number of MLIPs for various systems, including for elemental systems such as carbon, silicon, phosphorus, and tungsten, as well as for multicomponent systems such as Ge2Sb2Te5 and austenitic stainless steel, Fe7Cr2Ni. == Equivariant graph neural networks == A significant limitation of early MPNNs was that they were not inherently equivariant to rotations and reflections of atomic structures — meaning predictions could change depending on how a molecule was oriented in space. Beginning around 2021, a new class of models addressed this by incorporating equivariance directly into the message-passing layers using spherical harmonics and irreducible representations. Notable examples include NequIP (2021), MACE (2022), and GemNet-OC (2022). These equivariant architectures proved substantially more data-efficient and accurate than their predecessors, and became the dominant paradigm for high-accuracy MLIPs. == Universal MLIPs and large-scale datasets == Early MLIPs were system-specific, trained on a few thousand structures of a single material. A major shift occurred with the creation of large, chemically diverse datasets enabling models that generalize across many elements, bonding environments, and application domains — so-called universal MLIPs. A key driver was the Open Catalyst Project (OC20, OC22), a collaboration between Meta AI (FAIR) and Carnegie Mellon University launched in 2020. OC20 comprises approximately 1.3 million DFT relaxations across 82 elements, designed to accelerate the discovery of catalysts for renewable energy applications. It was among the first datasets large enough to train GNNs that generalize across diverse chemical systems, and established a widely-used benchmark for the field. A subsequent dataset, Open Direct Air Capture (OpenDAC 2023 and OpenDAC 2025), applied the same approach to carbon capture, providing a large computational database of metal-organic frameworks and sorbent candidates evaluated for CO₂ capture, generated using nearly 400 million CPU hours of quantum chemistry calculations in collaboration with Georgia Tech. These datasets revealed a new challenge: the GNN architectures most effective for atomic simulations were memory-intensive, as they model higher-order interactions between triplets or quadruplets of atoms, making it difficult to scale model size. Graph Parallelism, introduced by Sriram et al. (ICLR 2022), addressed this by distributing a single input graph across multiple GPUs — a distinct strategy from data parallelism (which distributes training examples) or model parallelism (which distributes layers). This enabled training GNNs with hundreds of millions to billions of parameters for the first time. Building on these foundations, Meta FAIR released the Universal Model for Atoms (UMA) in 2025, trained on approximately 500 million unique 3D atomic structures spanning molecules, materials, and catalysts — the largest training run to date for an MLIP. UMA introduced a Mixture of Linear Experts (MoLE) architecture, enabling one model to learn from datasets generated by different DFT codes and settings without significant inference overhead. It matches or surpasses specialized models across catalysis, materials, and molecular benchmarks without task-specific fine-tuning, and has been described as marking a "pre/post-UMA" divide in the field. == Applications == Catalyst discovery: MLIPs have significantly accelerated the computational screening of heterogeneous catalysts by replacing expensive DFT relaxations with fast neural network surrogates. The Open Catalyst Project explicitly targets this application, aiming to identify new catalysts for green hydrogen production and other renewable energy reactions. Carbon capture: The OpenDAC project applies universal MLIPs to screening sorbent materials for direct air capture of CO₂, a key technology for climate change mitigation. AI-accelerated screening allows evaluation of orders of magnitude more candidate materials than traditional DFT workflows. Drug discovery and molecular design: MLIPs are increasingly used in pharmaceutical research to model molecular conformations and binding energies. The Open Molecules 2025 (OMol25) dataset, released by Meta FAIR in 2025, provides high-accuracy calculations for a large set of molecular systems to support this use case. Materials discovery: Universal MLIPs enable high-throughput screening of novel inorganic materials, including battery electrolytes, semiconductors, and superconductors, by rapidly estimating stability and properties across large chemical spaces.
Aikuma
Aikuma is an Android app for collecting speech recordings with time-aligned translations. The app includes a text-free interface for consecutive interpretation, designed for users who are not literate. The Aikuma won Grand Prize in the Open Source Software World Challenge (2013). == Name == Aikuma means "meeting place" in Usarufa, a Papuan language where this software was first used in 2012. == History == Aikuma was developed with sponsorship from the National Science Foundation, including a $101,501 (US) project, "to use mobile telephones to collect larger amounts of data on undocumented endangered languages than would never be possible through usual fieldwork." Aikuma and its modified version (Lig-Aikuma) have been used for collecting substantial quantities of audio in remote indigenous villages. A modified version of the app, called Lig-Aikuma, has been developed at the Université Grenoble Alpes (LIG laboratory) and implements new features such as elicitation of speech from text, images and videos. == Similar Software == Lingua Libre is an online collaborative project and tool by the Wikimedia France association, which can be used as a tool for Language Preservation. Lingua Libre enables to record words, phrases, or sentences of any language, oral (audio recording) or signed (video recording). It is a highly efficient method to record endangered languages since up to 1000 words can be recorded per hour. All the content is under Free License, and speakers of minority languages are encouraged to record their own dialects.
Polyworld
Polyworld is a cross-platform (Linux, Mac OS X) program written by Larry Yaeger to evolve Artificial Intelligence through natural selection and evolutionary algorithms. It uses the Qt graphics toolkit and OpenGL to display a graphical environment in which a population of trapezoid agents search for food, mate, have offspring, and prey on each other. The population is typically only in the hundreds, as each individual is rather complex and the environment consumes considerable computer resources. The graphical environment is necessary since the individuals actually move around the 2-D plane and must be able to "see." Since some basic abilities, like eating carcasses or randomly generated food, seeing other individuals, mating or fighting with them, etc., are possible, a number of interesting behaviours have been observed to spontaneously arise after prolonged evolution, such as cannibalism, predators and prey, and mimicry. Each individual makes decisions based on a neural net using Hebbian learning; the neural net is derived from each individual's genome. The genome does not merely specify the wiring of the neural nets, but also determines their size, speed, color, mutation rate and a number of other factors. The genome is randomly mutated at a set probability, which are also changed in descendant organisms.
Fuzzy classification
Fuzzy classification is the process of grouping elements into fuzzy sets whose membership functions are defined by the truth value of a fuzzy propositional function. A fuzzy propositional function is analogous to an expression containing one or more variables, such that when values are assigned to these variables, the expression becomes a fuzzy proposition. Accordingly, fuzzy classification is the process of grouping individuals having the same characteristics into a fuzzy set. A fuzzy classification corresponds to a membership function μ C ~ : P F ~ × U → T ~ {\textstyle \mu _{\tilde {C}}:{\tilde {PF}}\times U\to {\tilde {T}}} that indicates the degree to which an individual i ∈ U {\textstyle i\in U} is a member of the fuzzy class C ~ {\textstyle {\tilde {C}}} , given its fuzzy classification predicate Π ~ C ~ ∈ P F ~ {\textstyle {\tilde {\Pi }}_{\tilde {C}}\in {\tilde {PF}}} . Here, T ~ {\textstyle {\tilde {T}}} is the set of fuzzy truth values, i.e., the unit interval [ 0 , 1 ] {\textstyle [0,1]} . The fuzzy classification predicate Π ~ C ~ ( i ) {\textstyle {\tilde {\Pi }}_{\tilde {C}}(i)} corresponds to the fuzzy restriction " i {\textstyle i} is a member of C ~ {\textstyle {\tilde {C}}} ". == Classification == Intuitively, a class is a set that is defined by a certain property, and all objects having that property are elements of that class. The process of classification evaluates for a given set of objects whether they fulfill the classification property, and consequentially are a member of the corresponding class. However, this intuitive concept has some logical subtleties that need clarification. A class logic is a logical system which supports set construction using logical predicates with the class operator { ⋅ | ⋅ } {\textstyle \{\cdot |\cdot \}} . A class C = { i | Π ( i ) } {\displaystyle C=\{i|\Pi (i)\}} is defined as a set C of individuals i satisfying a classification predicate Π which is a propositional function. The domain of the class operator { .| .} is the set of variables V and the set of propositional functions PF, and the range is the powerset of this universe P(U) that is, the set of possible subsets: { ⋅ | ⋅ } : V × P F → P ( U ) {\displaystyle \{\cdot |\cdot \}:V\times PF\rightarrow P(U)} Here is an explanation of the logical elements that constitute this definition: An individual is a real object of reference. A universe of discourse is the set of all possible individuals considered. A variable V :→ R {\textstyle V:\rightarrow R} is a function which maps into a predefined range R without any given function arguments: a zero-place function. A propositional function is "an expression containing one or more undetermined constituents, such that, when values are assigned to these constituents, the expression becomes a proposition". In contrast, classification is the process of grouping individuals having the same characteristics into a set. A classification corresponds to a membership function μ that indicates whether an individual is a member of a class, given its classification predicate Π. μ : P F × U → T {\displaystyle \mu :PF\times U\rightarrow T} The membership function maps from the set of propositional functions PF and the universe of discourse U into the set of truth values T. The membership μ of individual i in Class C is defined by the truth value τ of the classification predicate Π. μ C ( i ) := τ ( Π ( i ) ) {\displaystyle \mu C(i):=\tau (\Pi (i))} In classical logic the truth values are certain. Therefore a classification is crisp, since the truth values are either exactly true or exactly false.
Jake Elwes
Jake Elwes () is a British media artist, hacker and researcher. Their practice is the exploration of artificial intelligence (AI), queer theory and technical biases. They are known for using AI to create art in mediums such as video, performance and installation. Elwes considers themselves to be neuroqueer, and their work on queering technology addresses issues caused by the normative biases of artificial intelligence. == Education and early life == Elwes was born in London to British contemporary artist and painter Luke Elwes and Anneke, daughter of Hans Dumoulin. Elwes is the great grandchild of Army officer James Hennessy and portrait painter Simon Elwes RA, son of Victorian opera singer Gervase Elwes. Elwes studied at the Slade School of Fine Art from 2013 to 2017, where they began using computer code as a medium. In 2016 they attended the School of Machines, Making & Make-Believe in Berlin with artist and educator Gene Kogan. Elwes was introduced to drag performance by their collaborator Dr Joe Parslow who holds a PhD in drag performance. Drag performance has since become instrumental to Elwes' work. == Career == Elwes' work with artificial intelligence is cited as a hopeful strategy to make AI more playful and diverse. Elwes' work has been exhibited in numerous international art museums and galleries and was featured in a BBC documentary on the history of video art, they were a 2021 finalist for the Lumen Prize, and received the Honorary Mention of the 2022 Prix Ars Electronica in the Interactive Art + category. They also curated and presented the opening provocation "The New Real - Artistic and Queer Visions of AI Futures" to the UK government with two drag artists at the AI UK conference 2024. Elwes is part of the Radical Faeries countercultural movement. They have exhibited in museums and galleries across Europe and Asia including: Victoria and Albert Museum (London, UK) - The Zizi Show (2023-2024) for the first digital commission in their photography center's digital gallery Pinakothek der Moderne (Munich, Germany) - Glitch. Die Kunst Der Störung (2023-2024) ZKM (Karlsruhe, Germany) - Biomedia (2021-2022) National Museum of Modern and Contemporary Art (Cheongju, South Korea) - What an Artificial World (2024) Somerset House (London, UK) - The Horror Show! (2022-2023) Gazelli Art House (London, UK) - Jake Elwes: Data • Glitch • Utopia (2023) (survey exhibition) Jut Art Museum (Taipei, Taiwan) - Future Lives, Future You (2023-2024) Max Ernst Museum (Brühl, Germany) - Surreal Futures (2023-2024) Zabludowicz Collection (London, UK) - Among the Machines (2022) Ars Electronica (Linz, Austria) - Prix Ars Electronica, CyberArts Exhibition (2022) Institute of Contemporary Arts (ICA) (London, UK) - Do Androids Dream on Silver Screens? (2023) Arebyte gallery (London, UK) - Real-Time Constraints (2020) Ming Contemporary Art Museum (McaM) (Shanghai, China) - Mind the Deep (2019) HMKV (Hartware MedienKunstVerein) (Dortmund, Germany) - House of Mirrors: Artificial Intelligence as Phantasm (2022) Today Art Museum (Beijing, China) - Future of Today: DEJA VU (2019) Science Gallery (Dublin, Ireland) - BIAS (2021-2022) Yuz Museum (Shanghai, China) - Lying Sophia and Mocking Alexa (2021) Fotomuseum Winterthur The Onassis Foundation (Athens, Greece) - You and AI (2021) Royal College of Art (London, UK) - Event Two (2019) (50th anniversary of Computer Arts Society & Event One) Museum für Naturkunde (Berlin, Germany) - Forschungsfall Nachtigall (2019) Frankfurter Kunstverein (Frankfurt, Germany) - I am here to learn (2018) Nature Morte (Delhi, India) - Gradient Descent (2018) BALTIC Centre for Contemporary Art (Newcastle, UK) - Bloomberg New Contemporaries (2017) == Artworks == === The Zizi Project - a deepfake drag cabaret === The Zizi Project is a series of works that explore the interaction of drag and A.I. Currently, The Zizi Project is made up of multiple artworks. ==== Zizi - Queering the Dataset (2019) ==== Knowing that facial recognition technology statically struggle to recognize black women or transgender people, Elwes set out to "Queer the Dataset" through an open-sourced generative adversarial network (GAN, a type of machine learning model and an early Generative artificial intelligence). Elwes added a dataset of 1,000 photos of drag kings and queens into the GAN's 70,000 faces collected in a standardised facial recognition dataset called Flickr-Faces-HQ Dataset (FFHQ). They then created new simulacra faces, known as deep fakes. "We queer that data so it shifts all of the weights in this neural network from a space of normativity into a space of queerness and otherness. Suddenly all of the faces start to break down and you see mascara dissolve into lipstick and blue eye shadow turn into a pink wig" said Elwes in a 2023 interview for Artnet. ==== Zizi & Me (2020–2023) ==== Zizi & Me is an ongoing multimedia collaboration between drag queen Me The Drag Queen and a deepfake A.I. clone of Me The Drag Queen. Using neural networks trained on filmed footage, the project creates a virtual body that can mimic reference movements. The first act, which features a digital lip-sync duet to Anything You Can Do (I Can Do Better), satirises the idea of A.I. being mistaken for a human, using drag performance and cabaret to critique societal narratives about A.I. and its role in shaping identity. The project is part of The Zizi Project by Jake Elwes, which explores the intersection of drag performance and A.I. ==== The Zizi Show - A Deepfake Drag Cabaret (2020) ==== The Zizi Show is a deep fake drag act based on artificial intelligence (AI). It has been presented live and as interactive online artwork. It is an exploration of queer culture and the algorithms philosophy and ethics of AI. The Zizi Show was exhibited as the inaugural exhibition in the digital gallery at the V&A’s Photography Center from 2023 to 2024. ==== Zizi in Motion: A Deepfake Drag Utopia (Movement by Wet Mess) (2023) ==== "Zizi in Motion" is a multichannel silent video installation featuring AI-generated deepfake performances, which are dynamically re-animated through the movements of London drag artist Wet Mess. The movements of Wet Mess cause the AI-generated visuals to glitch and distort, showcasing the interaction between drag performance and artificial intelligence. The work explore the potential for queer communities to ethically and creatively reclaim and repurpose deepfake technology, using it to celebrate queer bodies and identities. === Art in the Cage of Digital Reproduction (2024) === In an act of protest on 26 November 2024, Elwes facilitated indirect access to an early access token for OpenAI’s Sora text-to-video model through a Hugging Face frontend under the account "PR Puppets". The accompanying statement called to 'denormalize the exploitation of artists by major AI companies for training data, R&D, and publicity'. The incident attracted international press coverage calling into question the role of artists in shaping the future of generative AI versus merely serving as data and credibility providers for tech giants. Elwes also coordinated a collection of mini essays with responses and reflections from the signees and guest writers titled "Art in the Cage of Digital Reproduction". === Installations exploring interpretation and feedback loops between neural networks === Elwes has created works based on the interpretations and misinterpretations between different neural networks and training datasets including: A.I. Interprets A.I. Interpreting ‘Against Interpretation’ (Sontag 1966) from 2023, Closed Loop from 2017, and Auto-Encoded Buddha from 2016. ==== A.I. Interprets A.I. Interpreting ‘Against Interpretation’ (Sontag 1966) (2023) ==== A.I. Interprets A.I. Interpreting ‘Against Interpretation (Sontag 1966) is a three-channel video artwork where an AI interprets Susan Sontag’s essay into images, and then and another AI reinterprets those images back into language. The piece highlights how AI-generated art can misinterpret and introduce bias. ==== Closed Loop (2017) ==== Closed Loop is a two-channel video where two neural networks engage in a continuous feedback loop, one generating images based on the text output and the other creating text based on the image output. The work explores how AI models misinterpret and evolve in a surreal, self-perpetuating conversation, without human input. ==== Auto-Encoded Buddha (2016) ==== Auto-Encoded Buddha is a mixed-media piece where an AI attempts to generate an image of a Buddha statue, trained on 5,000 Buddha images. The AI struggles to accurately represent the Buddha, highlighting the limitations of early generative neural networks. The work is a tribute to Nam June Paik’s TV Buddha (1974). === CUSP (2019) === In their video work CUSP (2019) Elwes places marsh birds generated using artificial intelligence into a tidal landscape. These digitally generated and constantly shifting birds are recorded in dialogue with native
Kruti
Kruti is a multilingual AI agent and chatbot developed by the Indian company Ola Krutrim. It is designed to perform real-world tasks for users, such as booking taxis and ordering food, by integrating directly with various online services. It is notable for its ability to understand and respond in multiple Indian languages. Developed by a team founded by Bhavish Aggarwal, Kruti functions as an "agentic" AI, meaning it can reason, plan, and execute multi-step tasks to fulfill a user's request. The backend technology combines several open-source large language models with Ola's proprietary Krutrim V2 model. The system was developed to work primarily on smartphones, addressing the Indian market's specific needs, including language diversity and potential bandwidth constraints. Kruti was officially released in June 2025, replacing an earlier chatbot from the company that was also named Krutrim. Initially supporting 13 languages, the company plans to expand its capabilities to 22 Indian languages. == Background == Kruti is an improved version of Ola's Krutrim chatbot, which was first launched in 2023 and was intended to be replaced by Kruti. It was officially released on 12 June 2025 as an upgrade to passive chatbots, with support for text and voice in 13 Indian languages. As an agentic AI, it can execute tasks with customization and reasoning, providing adaptive answers based on user preferences and past interactions. Kruti is optimized for smartphone usage and designed to accommodate bandwidth constraints and usage patterns in India. To ensure scalability and cost-effective performance, it combines various open-source large language models with Ola's own Krutrim V2, which has 12 billion parameters. Its speech recognition is built to identify regional Indian languages, dialects, and accents. Due to its integration with numerous apps and services, Kruti is context-aware and can proactively complete tasks. Initially connected only with Ola ecosystem services, Krutrim intends to expand and incorporate various Indian services into Kruti, with the goal of adding services from Blinkit, Swiggy, and Uber with respective voice command support. On 20 June 2025, Krutrim acquired the AI platform BharatSah‘AI’yak to increase its involvement in government, education, and agriculture projects. This acquisition will allow Kruti to assist in broadening the scope of BharatSah'AI'yak's work on India-centric, vernacular retrieval-augmented generation AI bots. == Development == Kruti is designed to perform tasks with minimal user input, accepting documents, images, and text, without requiring users to switch between applications. Its agentic framework breaks queries into sub-tasks executed by multiple agents working sequentially or concurrently, with reported accuracy exceeding 90%. Kruti connects to company databases and APIs via the Model Context Protocol and presents responses as summaries, tables, or narratives adapted to user behaviour. The system supports payments via credit/debit cards and UPI. The underlying stack, which includes foundation models and AI training and inference systems, is intended to support adaptation across sectors such as healthcare, education, and finance. Ola Cabs and the Open Network for Digital Commerce have begun integrating Kruti into their platforms pending broader reliability testing.
Resistance Database Initiative
HIV Resistance Response Database Initiative (RDI) was formed in 2002 to use artificial intelligence (AI) to predict how patients will respond to HIV drugs using data from more 250,000 patients from around 50 countries around the world. The RDI used its models to power its HIV Treatment Response Prediction System (HIV-TRePS). Launched in 2010, this free online tool enabled healthcare professionals to upload their patient’s data and obtain highly accurate predictions of how they would respond to different combinations of the 30 or more drugs available. The tool enabled physicians to individualize their patients’ treatment, using these predictions based on more than a million patient-years of treatment experience. HIV-TRePS was possibly the first ever AI-based system for medical decision-making to be developed, successfully tested, and used in clinical practice. It has since been used by thousands of healthcare professionals to optimise the treatment of tens of thousands of patients. Since the RDI’s inception the treatment of HIV infection has progressed enormously, with more effective and better tolerated drugs available in ever more convenient combination formulations. In most countries HIV is now considered a chronic, manageable condition. Moreover, the success of the drugs in reducing the amount of virus is substantially reducing the onward transmission of the virus and cases of new infections are falling in many settings. This improvement in HIV treatment means the need for sophisticated AI to support HIV treatment decisions has significantly reduced. In response, the RDI ceased development of further models and, in March 2024, withdrew its HIV-TRePS system. == Background == Human immunodeficiency virus (HIV) is the virus that causes acquired immunodeficiency syndrome (AIDS), a condition in which the immune system begins to fail, leading to life-threatening opportunistic infections. There are approximately 30 HIV antiretroviral drugs that have been approved for the treatment of HIV infection, from six different classes, based on the point in the HIV life-cycle at which they act. They are used in combination; typically 3 or more drugs from 2 or more different classes, a form of therapy known as highly active antiretroviral therapy (HAART). The aim of therapy is to suppress the virus to very low, ideally undetectable, levels in the blood. This prevents the virus from depleting the immune cells that it preferentially attacks CD4 cells and prevents or delays illness and death. Despite the expanding availability of these drugs and the impact of their use, treatments continue to fail, often involving to the development of resistance. During drug therapy, low-level virus replication may still occur, particularly when a patient misses a dose. HIV makes errors in copying its genetic material and, if a mutation makes the virus resistant to one or more of the drugs in the patient's treatment, it may begin to replicate more successfully in the presence of that drug and undermine the effect of the treatment. If this happens, the treatment needs to be changed to re-establish control over the virus. == RDI's Approach == The RDI’s approach was to use artificial intelligence (including neural network and random forest models), trained with data from hundreds of thousands of patients, treated with different drugs in a variety of clinical settings all over the world, to predict how an individual patient will respond to any new combination of HIV drugs. The models were tested with independent data sets and consistently achieved accuracy of approximately 80%.