China
Africa
Explore chapters and articles related to this topic
AI and Immunology Considerations in Pandemics and SARS-CoV-2 COVID-19
Published in Louis J. Catania, AI for Immunology, 2021
AI and immunoinformatics are being used to better understand the structure of proteins involved in SARS-Cov-2 infection in search for potential treatments and vaccines. Proteins have a three-dimensional structure, which is determined by their genetically encoded amino acid sequence (next-gen sequencing of genetic code), and this structure influences the role and function of the protein. An AI Google DeepMind system called AlphaFold70 uses amino acid sequencing and protein structure to make predictions to construct a “potential of mean force” which can be used to characterize the protein’s shape. This system has been applied to predict the structures of six proteins related to SARS-CoV-2.71
Introduction to Artificial Intelligence and Deep Learning with a Case Study in Analyzing Electronic Health Records for Drug Development
Published in Harry Yang, Binbing Yu, Real-World Evidence in Drug Development and Evaluation, 2021
After the success of AI in computer vision and strategy games, AI researchers and domain knowledge experts started to collaborate on the application of AI in drug development. The first identified area is drug discovery, where the volume of big data such as cellular images, cell line data and omics data keep increasing and challenge the traditional analytic methods (Chen et al. 2018). Since the first AI breakthroughs in computer vision, cellular images are the first among multiple modalities of data to be analyzed using AI methods (Chen et al. 2016). Deep learning methods have been used to predict biologic properties of compounds based on single high-throughput imaging assays (Simm et al 2018), reducing the cost of drug development. Deep learning methods also have been used in image embeddings for drug repurposing (Victors, n.d.), which removes the noises caused by batch processing cellular images and enhances the signal for drug repurposing. Recently, DeepMind has developed a novel AI, AlphaFold, to predict 3D protein structures using genomic data (Evans et al. 2018). In 2018, they won first place in the 13th Critical Assessment of Structure Prediction (CASP) competition, which is an international protein folding prediction competition. Able to predict the 3D structure of proteins can play a fundamental role in drug discovery for diseases believed to be caused by misfolded proteins, such as Alzheimer's, Parkinson's, Huntington's, and cystic fibrosis (Evans et al. 2018).
Introduction
Published in Wei Zhang, Fangrong Yan, Feng Chen, Shein-Chung Chow, Advanced Statistics in Regulatory Critical Clinical Initiatives, 2022
Wei Zhang, Fangrong Yan, Feng Chen, Shein-Chung Chow
For implementation of the 21st Century Cures Act, the FDA has kicked off several critical clinical initiatives to accelerate medical product development and bring innovations and advances to patients. Among these, AI and ML are one of the most important pieces. In recent years, AI/ML has brought significant impacts for the drug discovery, development, manufacture and commercialization. Target identification and molecule designs are the two key topics in drug discovery. Machine learning methods have shown to be efficient to identify a new drug target, such as identifying Baricitinib as a potential treatment for COVID-19 in early pandemic and end up with an FDA-approved indication (Richardson et al. 2020). Recently, DeepMind released AlphaFold 2 has significantly improved protein structure prediction and achieved experimental level accuracy (Tunyasuvunakool et al. 2021). The solutions will have a profound impact on how molecules will be designed and optimized. Beyond drug discovery space, AI/ML methods also have had significant impacts on drug development, such as speeding up clinical enrollment through identifying promising sites, identifying the right sub population for precision medicine. Digital health, by leveraging digital device, has achieved significant attention recently. Apple, Eli Lilly and Evidation have demonstrated that digital device can identify Alzheimer's disease and cognitive impairment (e.g., https://cacm.acm.org/news/238766-apple-eli-lilly-evidation-present-first-results-from-digital-alzheimers-study/fulltext). Drug formulation and manufacturing also start to leverage AI/ML solutions to better manufacturing drugs and optimize supply chains. Among all the new technology in AI/ML, reinforcement learning and deep learning are the two key themes to drive innovations. We are going to provide additional details in Chapter 6.
Blueprint for antibody biologics developability
Published in mAbs, 2023
Carl Mieczkowski, Xuejin Zhang, Dana Lee, Khanh Nguyen, Wei Lv, Yanling Wang, Yue Zhang, Jackie Way, Jean-Michel Gries
Expanding upon this is the use of machine learning (ML) and artificial intelligence (AI) to predict developability properties, such as viscosity and aggregation propensity.201,202 Sequences, structures and molecular dynamics have been used to train machine learning algorithms, which then attempt to predict molecular properties.203 Recent examples include the use of ML to predict affinity and specificity,204 deamidation,205 methionine oxidation,206 hydrophobicity,19 and viscosity.207 Recently, AI has been explored to reveal optimal antibody and protein formulations by predicting protein-solvent interactions.208 In terms of protein structure prediction, AlphaFold and AlphaFold2 have been recently launched commercially to predict protein structures and even their complexes.209,210 ABlooper (Antibody “Looper”) was developed as a deep learning tool to readily predict CDR structures from antibody sequences.211 While their use may not be widespread in the pharmaceutical industry, the emergence of AI/ML may become routine as part of initial in silico efforts to screen and assess molecular properties and interactions prior to any experimental efforts.
Deciphering cross-species reactivity of LAMP-1 antibodies using deep mutational epitope mapping and AlphaFold
Published in mAbs, 2023
Tiphanie Pruvost, Magali Mathieu, Steven Dubois, Bernard Maillère, Emmanuelle Vigne, Hervé Nozach
Epitopes can be mapped by various experimental processes.2 Over the years, a wide range of techniques have been used to determine which areas of the antigens are recognized by the antibodies. These include structural methods,3 peptide-based approaches,4 mutagenesis methods5,6 and mass spectrometry.2,7 More recently, computational modeling has enabled prediction of the antigen/antibody interface.8,9 The field of protein structure prediction has seen unprecedented progress, notably with AlphaFold and RoseTTAFold.10 X-ray crystallography and more recently cryogenic electron microscopy (cryo-EM) are still considered as gold standards for providing precise information on interaction sites with near atomic resolution. More precisely, 3D structures of complexes of antibodies with their antigens reveal amino acids from both sides of the interacting partners (namely structural epitope for the antigen and structural paratope for the antibody) that are close to each other and the chemical bonds that contribute to stability of the complex. However, the exact role of each amino acid present in the interacting surface can be difficult to decipher. Indeed, not all amino acids within a 4–4.5 Å radius from the other partner are necessarily important contributors to the binding free energy or to the specificity of the interaction.2
Promiscuity in drug discovery on the verge of the structural revolution: recent advances and future chances
Published in Expert Opinion on Drug Discovery, 2023
Sarah Naomi Bolz, Michael Schroeder
However, all these approaches rely on a precise reproduction of the binding site and are challenged by intrinsic limitations of AlphaFold [2,121]. First and most importantly, AlphaFold predicts a single structural model for a given amino acid sequence, which is unable to capture the dynamics and conformational diversity of proteins. For instance, a kinase may appear in an active or inactive state and a channel can be open or closed. Such conformational differences are often associated with post-translational modifications and can substantially affect the properties of binding sites. In addition, the shape of the binding site can change upon ligand binding, referred to as induced fit. Second, binding sites can be positioned at interfaces of protein–protein or protein–nucleic acid complexes [2]. Although AlphaFold currently only predicts the structures of individual amino acid sequences, progress has been made to predict multimers as well [3]. Finally, binding sites not only consist of amino acids, but also include metals, organic cofactors, and water molecules, all of which may be relevant for ligand interaction. AlphaFold models do not contain these molecule types, but efforts have already been made to add cofactors and metals [122].