On October 2, the third edition of Insilico Medicine’s Pharma.AI Quarterly Launch webinar—titled “Towards Pharmaceutical Superintelligence” —was successfully closed with around 300 registrants all over the world from universities, healthcare institutions, international pharmaceutical companies, and the innovative biotech industry. During the webinar, Insilico software team illustrated the most up-to-date abilities of the Pharma.AI platform while presenting live demos as well as case studies.

The highlights are summarized as below:
In the field of biology exploration, PandaOmics the cloud-based omics and text data analysis engine was updated all the way from dataset to ranking score and output processing, introducing four novel LLM scores focusing on confidence, commercial tractability, druggability and mechanism clarity for more comprehensive prioritization decisions.
Based on PandaOmics' experience in target discovery and indication exploration, Insilico Medicine recently submitted a systematic model performance evaluation framework research paper to a preprint server, proposing TID-Pro—a platform that combines machine learning target identification with comprehensive benchmark evaluation—and TargetBench 1.0, a target identification benchmark evaluation system. This integrated approach provides an efficient pathway for evaluating target discovery models, helping to optimize and improve the quality and efficiency of target discovery—a critical step in drug discovery—through AI-based multi-omics and computational modeling methods.
Also, Generative Biologics for novel biomolecules including antibodies and peptides, has further streamlined the model training, filtering and validation processes, so that more detailed structure understanding with more predictive models could lead to better balanced AI-designed molecules. Moreover, the Insilico team shared an internal test case using the platform for rapid design of GLP1R-targeting peptide molecules: within 72 hours, Generative Biologics generated over 5,000 novel peptide molecules. The team screened 20 high-potential candidates based on predicted affinity scores and computational binding energy, with 14 molecules showing biological activity, including 3 that demonstrated highly effective single-digit nanomolar activity.
As for chemistry, Insilico’s proprietary Chemistry42 is now composed of 7 distinctive applications spanning molecular generation, free energy binding prediction, ADMET property prediction, kinase selectivity prediction, and retrosynthesis route screening, and able to produce more than 2,400 molecule candidates within dozens of hours, joining the flexibility of generative AI and the precision of physics-based methods. Moreover, MDFlow for molecular dynamics (MD) simulation was introduced as a brand-new application, and Nach01, the multimodal natural & chemical languages foundation model is currently available on AWS marketplace.
Beyond specialized drug development models, Insilico Medicine continues to expand its Pharma.AI scientific assistant models. Among these, Science42: DORA achieves automated document generation through the integration of multiple AI agents and recently underwent an AI core model upgrade, bringing enhanced reasoning capabilities and real-time content verification features, enabling connectivity and interoperability with more AI tools. Insilico Medicine's proprietary life model PreciousGPT series was recently featured in two peer-reviewed papers in the journal Aging, exploring aging-related biological mechanisms.

To obtain a systematic view of Pharma.AI, the generative AI-driven solution for drug discovery and more cutting-edge research, please refer to the following answers provided by the Insilico AI team.

PandaOmics
Why do we recommend researchers to use PandaOmics?
It brings together multi-omics data, literature mining, disease context, and biomarker insights - all ranked and scored by AI models trained on real-world and experimental datasets. It drastically cuts down the early R&D time and lets you generate and test novel hypotheses fast, even with minimal bioinformatics support.
How is AI implemented in PandaOmics?
As the volume of biomedical data is too massive for humans alone, AI helps prioritize what matters faster so that scientists can focus on the most promising drug targets and reduce R&D guesswork.

AI powers four key areas in Pandaomics:

• Reading scientific literature: AI scans papers to extract genes, diseases, and their relationships.
• Target ranking: Graph-based AI models rank potential drug targets based on their connections in biology.
• Trend detection: AI spots early signals, like a rise in interest or new trials for a target-disease pair.
• Report generation: Large language models create expert-level summaries on genes and their drug potential.

What are the distinctive advantages of PandaOmics?

• Depth of Omics Integration: Combines transcriptomics, proteomics, epigenetics, clinical trials and real-world evidence to prioritize targets with disease-specific confidence scores.
• Disease-focused Knowledge Graph: Pre-built ontologies linking genes, pathways, and indications allow “one-click” hypothesis generation.
• Bibliomics & Patent Analytics: Tracks novelty and competitive landscape within the same interface - few rivals integrate IP data as deeply.
• AI Transparency: Provides contribution heat-maps showing which data layers drove target ranking, aiding scientists' trust.
• Speed: Cloud inference returns ranked target lists in minutes, outperforming manual bioinformatic workflows that take days.

What’s the future of PandaOmics?
It will evolve from a discovery platform into a full target-to-candidate system, closely integrated with chemistry design tools and patient stratification models. With more real-world data, it will offer even more personalized and disease-specific insights, making drug development faster, cheaper, and more precise.

Geneartaive Biologics
Why do we recommend researchers to use Generative Biologics?
It’s not just another sequence generator. It lets you generate, score, and optimize candidates with multi-model AI pipelines, and even retrain them on your own data. It’s built for real-world use: fast, flexible, and customizable.
How is AI implemented in Generative Biologics?
Generative Biologics uses a multi-model AI system where each model has unique capabilities:

• Large language models (LLMs) generate and optimize sequences using protein language understanding gained from the analysis of hundreds of millions of proteins.
• Graph Neural Networks (GNNs) learn 3D structural relationships for tasks like modeling protein-protein interactions.
• Diffusion models are used to understand protein shapes and design binder scaffolds that can interact with the desired target.
• AI predictors are based on complex models but solve highly specific tasks like predicting candidate affinity or developability properties.
• In Generative Biologics, AI is trainable, which means the models can evolve and improve their performance by receiving feedback from wet-lab data.

What are the distinctive advantages of Generative Biologics?
Unlike many tools that just generate sequences, Generative Biologics goes further. It scores and ranks generated molecules using both classical physics-based and AI-driven methods. It allows you to retrain models on your proprietary data to obtain even better predictive power for your projects. It supports multiple biologic types, offers multi-parameter optimization, and doesn’t require structural data for every use case. Plus, it’s designed for flexibility and scalability, making it usable by teams with different experience levels.
What’s the future of Generative Biologics?
Today, it supports peptides, antibodies, and nanobodies. Soon, it will expand into more complex modalities like bispecifics, antibody-drug conjugates, enzymes, and general protein therapeutics – broadening its impact across drug discovery.
We're also building foundational models that require minimal input data, allowing users to begin optimizing from a single sequence – no pretraining required. Integrating epitope mapping, clustering, and diverse candidate selection will further accelerate early-stage discovery.
Long term, Generative Biologics could potentially become a core AI assistant for biologics R&D – one that not only generates and ranks candidates, but also continuously learns from user data, integrates with wet-lab systems, and guides teams from hit discovery to preclinical development in record time.

Chemistry42
Why do we recommend researchers to use Chemistry42?
It goes beyond basic molecule generation, offering a comprehensive AI suite with powerful generative models for de novo design, multi-parameter optimization, ADMET profiling, retrosynthesis, and physics-based simulations. You can also train models on your proprietary data for customized workflows. It has achieved notable successes, such as advancing TNIK inhibitors to clinical stages in just 18 months, and it's highly adaptable for pharmaceutical or biotechnology teams aiming to reduce timelines and enhance innovation.
How is AI implemented in Chemistry42?
Chemistry42 implements AI through a multi-model system tailored for small-molecule drug discovery, where each component addresses specific aspects of molecular design and optimization.

• Generative models create novel molecules based on user-defined criteria, trained to excel on diverse chemotypes for reliable experimental outcomes.
• Retrosynthesis models, trained on expert-annotated reaction templates, predict synthetic routes with considerations for chemo-, regio-, and stereo-selectivity, integrated with vast libraries of building blocks.
• Property profiling leverages AI predictors to forecast and optimize DMPK parameters, toxicity profiles, and potency-related properties, available as a standalone tool or integrated into generative workflows.
• Golden Cubes utilizes self-organizing maps for estimating off-target kinome selectivity, working with 2D and 3D structures.
• The platform is trainable, allowing users to fine-tune models on proprietary data like in vitro activity or simulations for customized performance.
• Chemistry42 GPT, an LLM-based assistant, navigates users across the platform, aiding in workflow configuration and query handling.

What are the distinctive advantages of Chemistry42?

• AI-First Generative Ensemble: Chemistry42 features an extensive ensemble of generative AI models coupled with reinforcement learning to refine designs based on predefined user criteria, ensuring reliable experimental confidence and enabling de novo molecule creation.
• Integration of Physics-Based Algorithms: Seamlessly combines AI with physics-based core methods for superior accuracy and speed, addressing limitations in pure AI systems like data dependency and handling infinite chemical diversity.
• Custom Model Training on User Data: Allows users to train predictive models using their own proprietary data, creating highly tailored solutions that adapt to specific projects, datasets, and therapeutic goals for enhanced relevance and accuracy.
• Retrosynthesis: Predict reliable synthetic routes for uploaded or generated molecules using expert-annotated reaction templates and an AI-powered route planner. Built on a curated library of medicinal chemistry reactions and 300K commercially available building blocks - with support for chemo-, regio-, and stereo-selectivity - this module delivers complete synthetic pathways and reported examples. Leverage custom building block collections to streamline the synthesis planning.
• Advanced Filtering and Scoring Mechanisms: Incorporates over 460 Medicinal Chemistry Filters (MCFs) to exclude undesirable structures (e.g., PAINS or reactive groups) and the unique Medicinal Chemistry Evolution (MCE-18) score for assessing molecular novelty based on sp³ complexity. Additionally, the Retrosynthesis Related Synthetic Accessibility (ReRSA) score improves feasibility estimates by incorporating commercially available building blocks, addressing common limitations in synthetic route prediction found in other tools.
• Flexible Deployment and Scalability: Available as a SaaS solution or deployable on cloud platforms like AWS or Azure, with support for external integrations (e.g., QSAR models), offering greater scalability and interoperability for enterprise workflows compared to more siloed competitors.
• Experimental Validation Emphasis: Capabilities are thoroughly validated through in vitro, in vivo, and clinical studies, with successes like advancing multiple AI-designed candidates (10 in trials, e.g., TNIK for IPF) in timelines as short as 30 months to complete Phase 1, far faster than traditional 3-6 years.

What’s the future of Chemistry42?
Currently focused on de novo design and optimization, it will expand to hybrid modalities such as PROTACs, degraders, and potentially larger molecules, while quarterly updates - building on recent 2025 launches - introduce enhanced features like foundational AI that requires minimal input data for rapid optimizations. Collaborations will proliferate, broadening applications to fields like agrochemistry and materials science. Long-term, by 2028, Chemistry42 could serve as the cornerstone AI co-pilot in pharmaceutical R&D, seamlessly integrating with wet-lab robotics for closed-loop "design-make-test-analyze" cycles, continuously learning from real-world and proprietary data to predict clinical outcomes earlier, and driving even more candidates into trials - potentially tripling current successes - while slashing development timelines and costs for personalized, innovative therapies.

Nach01
Why do we recommend researchers to use Nach01?
It combines NLP with advanced 2D (like SMILES) and 3D point cloud processing to handle everything from de novo molecule generation and ADMET forecasting to hit discovery and lead optimization, all while allowing you to fine-tune it on your proprietary datasets for customized accuracy. As an "oracle" in reinforcement learning workflows, it boosts experimental success rates, and it's easily deployable via AWS Marketplace for seamless integration. Building on the proven Nach0 family, it's versatile for pharma or biotech teams looking to accelerate R&D without the limitations of single-task models.
How is AI implemented in Nach01?
Nach01 implements AI through a multimodal large language model (LLM) architecture designed as an encoder-decoder framework, pre-trained on massive datasets encompassing scientific literature, patents, and molecular representations to infuse deep chemical and linguistic knowledge.

• Multimodal Encoders: It employs a domain-specific encoder for processing textual data (e.g., natural language queries or chemical strings like SMILES) and a molecular point cloud encoder for handling spatial 3D data, enabling the model to understand atomic arrangements and overcome limitations of string-based representations.
• Generative and Predictive Capabilities: Leverages generative techniques to create novel molecules or textual outputs, while predictive modules forecast properties such as ADMET or activity, supporting tasks like de novo design, hit discovery, and lead optimization.
• Instruction and Multi-Task Fine-Tuning: The model is refined through instruction tuning for versatile task handling (e.g., biomedical QA, named entity recognition) and multi-task fine-tuning to excel across domains, ensuring flexibility in single- or cross-domain applications.
• User-Driven Customization: Allows fine-tuning on proprietary datasets, adapting the model for specific project needs and embedding it as an "oracle" in reinforcement learning workflows to refine designs iteratively.

What are the distinctive advantages of Nach01?

• Pioneering Availability: To our knowledge, Nach01 is the first multimodal generative chemistry foundation model available on major marketplaces like AWS Marketplace (launched in June 2025), making it easily accessible for deployment, inference, and fine-tuning without the need for in-house infrastructure, unlike many proprietary or academic models that require custom setups.
• Multimodal Processing: Seamlessly integrates textual (NLP), structural (2D like SMILES), and spatial (3D point clouds) data in a single encoder-decoder LLM framework, addressing limitations of string-based models that overlook 3D atomic arrangements or require separate pipelines for spatial tasks.
• Versatile Multi-Task Capabilities: Handles a broad range of chemistry tasks - from de novo molecule generation to property predictions (e.g., ADMET, activity) - in one model, offering greater flexibility than single-task tools.
• User-Driven Fine-Tuning: Enables easy customization by fine-tuning on proprietary datasets, creating tailored models for specific projects (e.g., hit discovery or lead optimization), providing an edge over rigid, non-adaptable models that can't incorporate user-specific data for improved relevance and precision.
• Scalable Deployment: Available via pay-as-you-go on AWS, supporting seamless integration into enterprise workflows without additional hardware, contrasting with open-source models (e.g., on Hugging Face) that may require extensive setup or lack commercial support for production-scale use.

What’s the future of Nach01?
In the future, Nach01 might be deeply integrated into multiple agentic workflows - leveraging AI agents for autonomous, multi-step processes like hypothesis generation, iterative molecule refinement, and experimental validation - empowering end-to-end drug discovery projects within Pharma.AI and custom pipelines, potentially increasing experimental success rates, reducing timelines from years to months, and allowing more candidates to reach clinical stages through ongoing learning from real-world data.

PreciousGPT
Why do we recommend researchers to use PreciousGPT?
When you have a hypothesis that costs too much to test in real life, e.g. in situations when you need to run a mass screening of chemical compounds. Before you allocate the funds and resources to such an endeavor, you can try asking one of the Precious models how a tissue or a cell would respond to your experiment. Its answer is by no means the final truth, but it gives you something to guide you to plan for the real-life execution of your experiment.
How is AI implemented in PreciousGPT?
PreciousGPT are a lineup of models. In each iteration we use different architectures to make sure we benefit from the most advance solutions in the overall deep learning field. Most Precious models are transformer-based, a type of architecture all current LLMs are based on. But unlike a chat bot, our models do not speak in a human language. They talk in genes and molecules, which means that we also need to invent new ways to represent these elements of life in a digital format.
What are the distinctive advantages of PreciousGPT?
Our models are developed with practical utility in mind using actual research data. Compared to regular LLMs, they are not biased toward only published research and can identify molecular mechanisms missed by human researchers. All Precious models are designed to model specific steps in the drug discovery pipelines, unlike other bio-AI models, which may show exciting functionality which is nonetheless not directly applicable.
What’s the future of PreciousGPT?
We aim to continue developing the Precious lineup of models by integrating even more biodata types, or even combining them with AIs from other domains, such as chemistry. We hope that eventually these models will evolve to the point when they could be used as a Life Model, or a realistic digital replica of any living system from a cellular to an organismal level.

About Insilico Medicine
Insilico Medicine, a leading and global AI-driven biotech company, utilizes its proprietary Pharma.AI platform and cutting-stage automated laboratory to accelerate drug discovery and advance innovations in life sciences research. By integrating AI and automation technologies and deep in-house drug discovery capabilities, Insilico is delivering innovative drug solutions for unmet needs including fibrosis, oncology, immunology, pain, and obesity and metabolic disorders. Additionally, Insilico is extending the reach of Pharma.AI across diverse industries, such as advanced materials, agriculture, nutritional products and veterinary medicine.
For more information, please visit www.insilico.com.