AI Retrosynthesis FAQ: CASP, Synthesizability & More

Answers on AI-driven retrosynthesis, CASP tools, and synthesizability scoring from Chemical.AI - covering how ChemAIRS® supports medicinal chemists, process development teams, and drug discovery organizations.

Understanding AI-Powered Retrosynthesis

  • Retrosynthesis is a problem-solving strategy in organic chemistry where a chemist works backwards from a complex target molecule, systematically breaking it down into simpler precursor molecules until commercially available starting materials are reached. First formalized by Nobel laureate E.J. Corey in the 1960s, retrosynthetic analysis is the foundation of modern synthesis planning. AI-powered platforms like ChemAIRS® automate this process, analyzing thousands of possible retrosynthetic pathways in minutes and ranking them by feasibility, cost, and risk.

  • AI-driven retrosynthesis uses machine learning and rule-based models to automate the retrosynthetic process, instead of relying on one chemist's intuition and experience. It analyzes thousands of possible bond disconnections, proposes reagents and conditions with links to literature references, and ranks routes by feasibility, cost, and risk. ChemAIRS® uses a hybrid approach combining AI and curated rules, which allows it to propose both established routes grounded in published chemistry and novel, chemically plausible pathways that have never been published.

  • CASP refers to software that uses algorithms and chemical databases to assist chemists in designing synthetic routes, reducing the time and effort required for manual literature searching and route brainstorming. ChemAIRS® is consistently ranked number one among leading CASP tools in independent evaluations for idea feasibility and diversity, and is trusted and adopted by the majority of the top 10 pharmaceutical companies, CROs, and CDMOs.

  • Retrosynthesis works backwards from a target molecule to identify simpler starting materials - it answers "how do we make this?" Forward synthesis works in the opposite direction, starting from available building blocks or a core scaffold and generating what could be made from them - it answers "what could we make?" ChemAIRS® supports both within a single platform; its Forward Synthesis module helps teams generate focused analogue libraries under defined chemistry constraints, using the building blocks they actually have access to.

  • FGI stands for Functional Group Interconversion - a strategy in retrosynthetic analysis where one functional group is transformed into another to make a strategic bond disconnection possible. It's used when a direct disconnection of the target molecule isn't immediately feasible. FGI is one of the strategies ChemAIRS® considers when generating retrosynthetic pathways.

  • The SA Score, or Synthesizability Assessment Score, evaluates how practical a molecule is to synthesize, helping chemists and computational teams prioritize which compounds from a virtual library are realistic to make before investing time and resources. In ChemAIRS®, the SA Score is calculated based on retrosynthesis rules rather than chemical similarity. The score runs from 0 to 5: 0 means the compound is commercially available, and the score increases as synthesis becomes longer and more complex. ChemAIRS® validated its SA Score against the assessments of a team of medicinal chemists at Merck across over 13,000 molecules, with close alignment.

  • No. A score of 5 means the algorithm did not find a straightforward route within the time available for the SA Score calculation - it does not mean the compound cannot be synthesized. Any compound with a score of 5 can be pushed directly into the Retrosynthesis module for deeper analysis, where routes are often found. SA Score is a prioritization tool, not a yes-or-no gate.

  • "Ideation Beyond Search" reflects that ChemAIRS® doesn't stop at what's already been published - it treats every target molecule as novel, generating chemically plausible routes whether or not literature precedent exists. As the platform puts it: "Where the literature stops, the thinking starts." This is what distinguishes route generation from a literature lookup: one returns what's known, the other proposes what's possible.

Evaluating ChemAIRS®

  • ChemAIRS® has been consistently ranked number 1 in idea feasibility and diversity in an independent evaluation conducted by a top chemistry synthesis CRO, where 9 leading CASP tools were evaluated across 60 diverse target molecules and scored by expert synthetic chemists. It is trusted and adopted by the majority of top 10 pharmaceutical companies, CROs, and CDMOs.

  • In an internal case study across 16 synthetic projects, ChemAIRS®' top 3 suggested routes aligned with the route an expert chemist independently chose in 80% of cases. Separately, in the independent CRO evaluation described above, ChemAIRS® ranked first among 9 leading CASP tools for both feasibility and diversity.

  • ChemAIRS® is 21 CFR Part 11 ready and uses AES-256 encryption, and supports full local, on-premise deployment for organizations with strict data security or compliance requirements. Every proposed reaction condition links to its original literature reference, and impurity predictions include a confidence score from 0-100 with a mapped reference reaction - giving a traceable basis for review rather than an unexplained output.

  • ChemAIRS® reduces synthesis planning time by approximately 90%. Tasks that traditionally take days are completed in minutes: literature search goes from 3-5 days to 30 seconds, route brainstorming from 2-3 days to 2-4 minutes, cost and reagent checking from 1-2 days to instant, and team review and ELN integration from 2-4 days to real time.

  • Yes. You can try ChemAIRS® directly on the Chemical.AI website by selecting "Try ChemAIRS®," or contact the team at contact@chemical.ai to get started.

ChemAIRS® Platform & Security

  • ChemAIRS® is Chemical.AI's retrosynthesis planning platform, trained on a reaction database of over 61 million reactions. It delivers ranked, cost-scored synthesis routes for any target molecule in minutes, across 8 modules: Retrosynthetic Analysis, Forward Synthesis, Synthesizability Assessment (SA Score), Impurity Prediction, Condition Optimization, Bayesian Optimization, Process Chemistry, and Internal Data Integration.

  • ChemAIRS® serves four main user groups. Drug discovery teams use it to explore thousands of pathways and reduce route planning from days to hours. Medicinal chemists use it for diverse, creative route design with real-time cost, yield, and condition prediction. Process development and scale-up teams use it to design cost-efficient, scalable routes from the start. Computational chemists use it to rank virtual libraries by practical synthesizability.

  • ChemAIRS® is available via subscription through any modern web browser. Chemical.AI also offers full local deployment for organizations with strict data security or compliance requirements. To get started, visit chemical.ai or contact contact@chemical.ai.

  • ChemAIRS® uses AES-256 encryption and supports full local deployment. It is 21 CFR Part 11 ready, making it suitable for regulated pharmaceutical environments. For organizations requiring on-premise deployment, all data stays within your own environment. ChemAIRS® also connects to existing lab systems via ELN integration, typically set up within days.

  • Yes. Internal Data Integration is one of the 8 core modules in ChemAIRS®. With local deployment, internal ELN data, building block libraries, and custom vendor catalogs can be connected, typically within days. Learnings are retained locally over time through adaptive learning.

Using ChemAIRS®

  • There are three ways: draw it directly using the built-in molecule editor, paste a SMILES code copied from ChemDraw or another editor, or use the AI Vision feature to screenshot a molecule from a paper or any other source and let the platform recognize the structure automatically.

  • First, increase the risk level in your search parameters - by default ChemAIRS® searches at a low risk level focused on well-established chemistry. Switching to a higher risk level allows exploration of more creative, novel routes. You can also manually break the molecule into smaller fragments using Manual Search, or run the search in the Process Chemistry module, which allocates more computational resources. If you're still not finding results, the Chemical.AI team can review your specific molecule directly.

  • ChemAIRS® ranks routes using a difficulty score - a weighted combination of the number of linear steps and how closely each proposed reaction resembles known, precedented chemistry. Routes can also be ranked with consideration of internal ELN data when local deployment is in place. ChemAIRS® delivers 10 to 50 prioritized routes per search, and in an internal case study, its top 3 routes aligned with expert chemist decisions in 80% of 16 synthetic projects.

  • Both modules generate synthetic routes but are optimized for different goals. Retrosynthesis is designed for discovery chemists and focuses on identifying the fastest, most feasible routes without significant concern for bulk material availability or large-scale cost. Process Chemistry allocates more computational resources - searches can take up to 4 hours - and prioritizes novelty, diversity, and scalability, with a preference for building blocks available in bulk. It flags conditions demonstrated at 100-gram scale or greater, or sourced from OPRD journals, to support scale-up decisions.

  • Yes. ChemAIRS®' Impurity Prediction module models a single-step reaction before it's run in the lab, using the reactants, conditions, and any known contaminants, including trace impurities from a previous step. Each predicted impurity is accompanied by a confidence score from 0 to 100, with the proposed reaction mechanism mapped to a real reference reaction. You can also upload a proton NMR spectrum and overlay it with predicted NMR spectra to confirm a structure before running more expensive analyses.

  • Yes. For simple chiral molecules with one or a few unrelated chiral centers, ChemAIRS® guides users on how the chirality can be constructed and provides relevant literature references. For molecules with multiple nearby chiral centers, it focuses on relative configuration with supporting references. Chiral separation and chiral auxiliaries functions have also been added, addressing absolute chiral configurations, with a dedicated chirality module in development for advanced process-chemistry use cases.

  • Yes. From any route, you can export a PDF or Word file with a full step-by-step breakdown, or generate a Compound List - a CSV of every building block, reagent, and solvent required - to cross-reference with inventory or pass to procurement. Routes can be shared with colleagues with one click, bookmarked, and commented on directly within the platform.

Data & Methodology

  • ChemAIRS® is trained on a reaction database of over 61 million reactions, combining data from patents including USPTO and EPO, open-source published literature, and Chemical.AI's own proprietary curated datasets. The platform uses a hybrid approach combining AI and curated rules, supporting both reliable coverage of known chemistry and predictive capability for novel molecules.

  • Yes. For each reaction step, ChemAIRS® proposes conditions - temperature, solvent, catalyst - and links directly to the original literature reference. The Condition Search tool gives a full overview of all available literature-backed conditions for that transformation, filterable by pressure, scale, reaction class, and risk. Conditions demonstrated at 100-gram scale or greater, or sourced from OPRD journals, are specifically flagged for process chemistry.

  • ChemAIRS® only includes building blocks that can ship within two weeks, so routes requiring materials with longer lead times may not appear. A published route may also be deprioritized if it scores higher in difficulty, cost, or number of steps than other routes the algorithm identified, or if the publication is too recent to have been incorporated into the database. ChemAIRS® is designed as a synthesis planning co-pilot, not a literature search tool - it analyzes every molecule as novel, making it complementary to dedicated chemical literature databases. If a specific published route is important to include, ChemAIRS® has a manual search and route import function that allows you to add it directly.

  • The ChemAIRS® software is updated approximately every three months. The AI model is retrained twice a year. UI fixes and bug patches are applied as needed. When local deployment is in place, internal ELN data continuously contributes to improving route recommendations over time through adaptive learning.

Contact Us