AI-Based Antibody Screening

AI-Augmented Drug Discovery Creative Biolabs provides innovative drug discovery services based on our original Artificial Intelligence-augmented technology, especially for the discovery of therapeutic antibodies and small molecules. Email: [email protected] Address: SUITE 203, 17 Ramsey Road, Shirley, NY 11967, USA Web: www.creative-biolabs.com

CONTENTS 01 02 03 04

01

WHY USE AI IN DRUG DISCOVERY? $2.6 B 10% 76% Introducing a new drug to market Even once new drug candidates Between 2010 and 2017, 76% of can cost pharmaceutical show potential in laboratory new drugs approved by the US testing, less than 10% of drug Food and Drug Administration companies an average $2.6 billion candidates make it to market and 11-15 years of research and following Phase I trials. (FDA) are small molecules. development.

AI in Drug Discovery AI in Clinical Trials (Phase I) (Phase III) The drug discovery process ranges from reading and analyzing After making it through the preclinical development already existing literature, to testing the ways potential drugs phase, and receiving approval from the FDA, interact with targets. According to report, AI could curb drug researchers begin testing the drug with human discovery costs for companies by as much as 70%. participants. AI can facilitate participant monitoring during clinical trials—generating a larger set of data AI in Preclinical Development more quickly—and aid in participant retention by (Phase II) personalizing the trial experience. The preclinical development phase of drug discovery involves testing potential drug targets on animal models. Utilizing AI during this phase could help trials run smoothly and enable researchers to more quickly and successfully predict how a drug might interact with the animal model.

AI In Drug Discovery AI in AI in AI in AI in AI in drug design polypharmacology chemical synthesis drug repurposing drug screening Ø Predicting 3D structure of Ø Designing biospecific Ø Predicting reaction yield Ø Identification of Ø Prediction of toxicity target protein drug molecules Ø Predicting retrosynthesis therapeutic target Ø Prediction of bioactivity Ø Prediction of Ø Predicting drug-protein Ø Designing multitarget pathways Ø Prediction of new interactions drug molecules Ø Developing insights into therapeutic use physicochemical property Ø Identification and Ø AI in determining drug reaction mechanisms activity Ø Designing synthetic route classification of target cells Ø AI in de novo drug design

02

Classes of Learning Tasks and Techniques Supervised Learning The goal is to reconstruct the unknown function f that assigns output values y to data points x. Unsupervised Learning Can be treated as a geometric or topological problem, the goal is to find similarities and differences between data points used to spatially order data. Semisupervised Learning (Fig. A) Mix of supervised and unsupervised learning, where less expensive and more abundant unlabeled data can be utilized to train a classifier. Active Learning (Fig. B) A learning algorithm can interactively query the user to determine labels for unlabeled data in the regions of the input space about which the model is least certain. Reinforcement Learning (Fig. C) To some extent strives to emulate reward-driven learning, and in its simplest configuration, an agent attempts to find the optimal set of actions to promote some outcome. Transfer Learning (Fig. D) Describes a family of algorithms that relax the common assumption that the training and test data should be in the same feature space and follow the same distribution. Multitask Learning (Fig. E) Instead of learning only one task at a time, as in single-task learning, several different but conceptually related tasks are learned in parallel and make use of a shared internal representation. Xin Y,et al. Concepts of Artificial Intelligence for Computer-Assisted Drug Discovery. Chem. Rev. 2019, 119 (18): 10520-10594.

Common Learning Algorithms Bayesian Algorithms Instance-Based Methods Decision Tree Algorithms Bayesian methods are those that explicitly It is called instance-based because it builds Algorithms for constructing decision trees apply Bayes’ theorem to classification and the hypotheses from the training instances. usually work top-down, by choosing a regression problems. It is also known as memory-based learning variable at each step that best splits the or lazy-learning. set of items. Ensemble Algorithms Dimensionality Reduction Artificial Neural Networks In statistics and machine learning, Dimensionality reduction seeks a lower- Artificial neural networks (ANNs) consist of ensemble methods use multiple dimensional representation of numerical input, hidden, and output layers with learning algorithms to obtain better input data that preserves the salient connected neurons (nodes) to simulate the predictive performance than could be relationships in the data. human brain. obtained from any of the constituent learning algorithms alone.

Bayesian Algorithms Liu ZH,et al. ChemStable: A web server for rule-embedded naïve Bayesian learning approach to predict compound stability. J. Comput. Aided Mol. Des. 2014, 28: 941-950.

Instance-Based Methods K-nearest Neighbor KNN is a simple, supervised machine learning algorithm that can be used to solve both classification and regression problems. Self-organizing Map A SOM or self-organizing feature map is an unsupervised machine learning technique used to produce a low-dimensional representation of a higher dimensional data set while preserving the topological structure of the data. Support Vector Machine SVM is a supervised machine learning algorithm used for both classification and regression. The objective of SVM algorithm is to find a hyperplane in an N-dimensional space that distinctly classifies the data points. Xin Y,et al. Concepts of Artificial Intelligence for Computer-Assisted Drug Discovery. Chem. Rev. 2019, 119 (18): 10520-10594.

Decision Tree Algorithms Decision Tree A decision tree is a decision support tool that uses a tree-like model of decisions and their possible consequences, including chance event outcomes, resource costs, and utility. Random Forest Random forests or random decision forests is an ensemble learning method for classification, regression and other tasks that operates by constructing a multitude of decision trees at training time. Xin Y,et al. Concepts of Artificial Intelligence for Computer-Assisted Drug Discovery. Chem. Rev. 2019, 119 (18): 10520-10594.

Ensemble Algorithms Bagging Boosting Bagging, is the ensemble learning method that is commonly used Boosting is an ensemble learning method that combines a set of weak to reduce variance within a noisy dataset. In bagging, a random learners into a strong learner to minimize training errors. In boosting, a sample of data in a training set is selected with replacement— random sample of data is selected, fitted with a model and then meaning that the individual data points can be chosen more than trained sequentially—that is, each model tries to compensate for the once. weaknesses of its predecessor. Xin Y,et al. Concepts of Artificial Intelligence for Computer-Assisted Drug Discovery. Chem. Rev. 2019, 119 (18): 10520-10594.

Dimensionality Reduction A visual depiction of the resulting PCA projection for a set of 2D points. A visual depiction of the resulting LDA projection for a set of 2D points. Principal Component Analysis Linear Discriminant Analysis PCA is a popular technique for analyzing large datasets containing a high LDA is a generalization of Fisher's linear discriminant, a method used in number of dimensions/features per observation, increasing the statistics, pattern recognition and machine learning to find a linear interpretability of data while preserving the maximum amount of information, combination of features that characterizes or separates two or more classes and enabling the visualization of multidimensional data. of objects or events. Image From Wikipedia

Artificial Neural Networks Artificial neural networks ANNs are computing systems inspired by the biological neural networks that constitute animal brains. A typical ANN architecture contains many artificial neurons arranged in a series of layers: the input layer, an output layer, i.e., the top layer, which generates a desired prediction ( ADMET properties, activity, a vector of fingerprint etc.), and one or more hidden layer where the intermediate representations of the input data are transformed. Deep Neural Networks DNN refers to an ANN that has several hidden layers with several differences. Deep nets process data in complex ways by employing sophisticated math modeling. Image From Wikipedia

03

DeepVS: Boosting Docking-Based Virtual Screening with DL DeepVS The deep neural network that is introduced, DeepVS, uses the output of a docking program and learns how to extract relevant features from basic data. The approach introduces the use of atom and amino acid embeddings and implements an effective way of creating distributed vector representations of protein–ligand complexes by modeling the compound as a set of atom contexts that is further processed by a convolutional layer. Pereira J.C. Boosting docking-based virtual screening with deep learning. J. Chem. Inf. Model. 2016;56:2495–2506. Mostafa K. DeepAffinity: interpretable deep learning of compound–protein affinity through unified recurrent and convolutional neural networks. Bioinformatics. 2019, 35(18):3329–3338.

DeepAffinity: DL Method Used to Measure DTBA DeepAffinity DeepAffinity is a deep learning methods used to measure drug target binding affinity. Under novel representations of structurally-annotated protein sequences, a semi-supervised deep learning model that unifies recurrent and convolutional neural networks has been proposed to exploit both unlabeled and labeled data, for jointly encoding molecular representations and predicting affinities. Performances for new protein classes with few labeled data are further improved by transfer learning. Mostafa K. DeepAffinity: interpretable deep learning of compound–protein affinity through unified recurrent and convolutional neural networks. Bioinformatics. 2019, 35(18):3329–3338.

DeepTox: Toxicity Prediction Using Deep Learning Representation of a toxicophore by hierarchically related features.  Mayr A. DeepTox: toxicity prediction using deep learning. Front. Environ. Sci. 2016, 3:80.

AI-Based QSAR Models Profile-QSAR Bayesian QSAR SVM QSAR Multitask QSAR Image From Wikipedia

04

AI-Based One-stop Antibody Discovery Platform Creative Biolabs has combined AI, big data, machine learning, and phage display to generate a novel AI-powered computational antibody drug discovery platform. Aided by this innovative platform, one-stop human antibody discovery services are provided, including antibody-antigen binding prediction, antibody candidate generation, antibody sequence optimization, and antibody production & characterization. Augmented Antibody Discovery with Al Antibody Production by Phage Display • Discover and analyze new antibody clusters • High throughput, screen large numbers of clones • Generate new sequences within existing clusters • Large library capacity: from 107 to over 108 • Accelerate the generation of high-affinity antibodies • Various phage display systems (M13,λ,T7) • Rapidly generate novel antibody sequences using • Tailored biopanning strategies • Wide range of applications computational algorithms to help improve affinity, solubility, manufacturability, specificity, and stability

AI-Augmented Drug Discovery at Creative Biolabs Antibody Discovery Services Antibody Engineering Services Model Training Data Services AI can typically generate 10 times more Creative Biolabs offers a wide variety of antibody Creative Biolabs provides the best strategy and antibody sequence clusters than a laboratory- engineering services to quickly and efficiently optimize customized protocols for model training data based approach alone. Diversity leads to the the existing antibodies via AI based algorithms, such service, and ultimately, to accelerate the novel discovery of new binding modalities and as affinity, solubility, cross-reactivity, manufacturability, candidate drug discovery. potentially new therapeutic modes-of-action. immunogenicity, specificity, and stability. Antibody Screening Services Small Molecule Design & Optimization Creative Biolabs is specialized in designing and Creative Biolabs has applied AI technology in small performing high-quality custom AI-based antibody molecule design and optimization to promote its affinity, screening assays, with different formats, endpoints, specificity, and validity. Our innovative AI methods range parameters, to satisfy any specific requirement. from in silico molecule screening, molecular modeling, to AI-based molecule optimization.

THANKS https://www.creative-biolabs.com Email: [email protected] Copyright © Creative Biolabs. All Rights Reserved.