Responsible Use of Generative AI in Medicinal Chemistry

A practical framework for helping young scientists improve productivity while safeguarding intellectual property and client confidentiality

Dr. Prasant Deb
Senior Vice President and Site Head, Noida and Greater Noida
Jubilant Biosys

Introduction

Generative AI is rapidly changing how scientists find information, explore ideas and communicate their work. In medicinal chemistry, it can help researchers understand unfamiliar concepts, investigate possible causes of experimental challenges and prepare clearer technical documentation. Yet in a contract research and drug discovery environment, its use must be balanced with an equally important responsibility: protecting intellectual property, client confidentiality and scientific integrity.

Many research organizations are evaluating structured approaches for the responsible use of generative AI in scientific environments. Over the past year, we have led an initiative to help young chemists to use approved generative AI tools effectively and responsibly to improve efficiency and productivity in their work.  The objective has never been to replace scientific expertise or judgement. It has been to help scientists learn faster, frame problems more clearly and reduce time spent on routine information gathering and drafting. Scientific conclusions, experimental decisions, client deliverables and research recommendations must always be generated, reviewed and approved by qualified scientists. AI-generated content should never be relied upon as a substitute for scientific judgment.

The foundation of the initiative is simple. Generative AI can support scientific thinking, but confidential science must never be entered into a public or unapproved AI system. This article outlines the practical framework, use cases and lessons that have guided our approach.

Why a structured approach is necessary

Medicinal chemists routinely work across reaction mechanisms, literature analysis, synthetic route planning, analytical interpretation, experimental troubleshooting and technical communication. Young scientists, in particular, may spend considerable time searching for foundational information, comparing methodologies and preparing summaries or presentations.

Generative AI can accelerate some of these activities. It can explain concepts in accessible language, suggest lines of enquiry and help structure technical content. However, unrestricted use can expose sensitive information, including compound structures, unpublished synthetic routes, structure-activity relationship data, experimental observations and client or project identifiers.

Responsible adoption therefore requires more than access to a tool. It requires clear boundaries, practical training and a culture in which every output is treated as a starting point for scientific evaluation, not as an authoritative answer.

The guiding principle

The principle shared with our young chemists is: use generative AI to improve your thinking, not to reveal your science.

Used within appropriate guardrails, generative AI can support scientific learning, literature understanding and general knowledge development including communication. It should never become a repository for confidential information, nor should it make decisions that require scientific accountability. As a rule of thumb, Client-owned information, data, inventions, research findings, compound structures, biological information, analytical results and project-related materials should never be entered into public AI platforms.

A three-part framework for responsible use

1. Protect confidential and proprietary information

Scientists must not enter client details, project identifiers, molecular structures, unpublished data, laboratory notebooks, batch records, proprietary reaction schemes or observations linked to active programmes into public or unapproved AI systems. Training should include realistic examples because individual details that appear harmless can, when combined, reveal a project or compound series. It is also imperative that use of AI tools does not alter ownership of intellectual property, confidentiality obligations or contractual commitments applicable to scientific work.

2. Generalise the scientific question

JUseful guidance can often be obtained without disclosing project-specific information. Instead of asking how to improve the yield of a named compound in a particular project, a scientist could ask: What factors commonly contribute to protodeboronation in a Suzuki coupling, and what general mitigation strategies are reported?

This approach moves the discussion from proprietary chemistry to mechanism, methodology and public-domain knowledge. It preserves confidentiality while still helping the scientist identify relevant variables to investigate.

3. Verify before applying

Generative AI outputs can be incomplete, outdated or incorrect. Every response must therefore be critically reviewed, checked against reliable literature and validated experimentally where relevant. The scientist remains responsible for assessing feasibility, safety, scientific relevance and compliance with established procedures. AI can inform a decision; it cannot own the decision.

Practical applications for young chemists

Reaction troubleshooting

When questions are framed generically, generative AI can help scientists explore possible reasons for low conversion, side-product formation or difficult purification. It may prompt consideration of catalyst choice, solvent effects, reagent quality, moisture sensitivity or competing reaction pathways. These suggestions are hypotheses to investigate, not instructions to follow without review.

Literature understanding and scientific learning

AI tools can help explain reaction mechanisms, compare established synthetic methodologies and provide an initial orientation to unfamiliar technologies. This is particularly useful when scientists encounter areas such as targeted protein degradation, lipid chemistry, peptide synthesis, photoredox chemistry, electrochemistry or flow chemistry. The tool can shorten the path to understanding, but primary literature and trusted databases remain the basis for scientific conclusions.

Technical writing and communication

Generative AI can help organise non-confidential content, improve clarity and adapt technical information for different audiences. Appropriate uses include refining generic experiment summaries, structuring meeting minutes, improving project-update language and developing presentation outlines. Scientists must review the final text for accuracy, context and tone, and must not upload confidential source documents to obtain a draft.

Root cause analysis and planning

For recurring laboratory challenges, AI can support an initial, structured exploration of possible causes, including sensitivity to moisture or oxygen, scale-up variables, mixing, heat transfer, impurity formation and safety considerations. This can broaden the questions considered before the team turns to literature review, experimental design and expert discussion.

Activities that remain outside the boundary

The framework explicitly prohibits uploading or sharing:

• Compound structures or active project chemistry
• Client names, project identifiers or confidential presentations
• Unpublished structure-activity relationship tables or biological data
• Experimental notebooks, raw datasets, spectra or confidential reports
• Proprietary synthetic routes, batch records or process information
• Requests to design chemistry around an active client programme

These boundaries must be reinforced through periodic training, practical examples and alignment with organisational information-security policies. Where there is any doubt, the information should not be entered.

What we have learned

Industry experience suggests, a structured approach may help young chemists become more confident in navigating unfamiliar scientific topics, broaden the range of hypotheses they consider and communicate technical information more clearly. It can also reduce effort in selected routine tasks, particularly early-stage information gathering and the organisation of non-confidential content.

The quality of adoption depends on the quality of training. Simply providing access to an AI tool is insufficient. Scientists need examples of acceptable and unacceptable prompts, a clear escalation route when they are uncertain, and repeated reminders that plausible language is not evidence of scientific accuracy.

Governance is equally important. Approved tools, access controls, organisational policies and human review should be defined before use is scaled. The framework must also evolve as AI capabilities, regulations and information-security expectations change.

Human expertise remains central

Successful chemistry depends on experimental insight, mechanistic understanding, critical thinking, creativity and experience. Generative AI does not reproduce the full context of a laboratory, understand the history of a programme or assume accountability for a scientific decision.

Its value lies in augmenting the scientist. The strongest users are not those who accept outputs quickly, but those who ask better questions, recognise uncertainty, verify claims and apply expert judgement.

Conclusion

Generative AI can be integrated responsibly into chemistry-led research when its role and boundaries are clearly defined. A practical model rests on three disciplines: protect confidential information, generalise scientific questions and verify every output before use. Organizations should ensure that any use of AI tools is consistent with applicable laws, contractual obligations, information-security requirements, intellectual property protections and industry regulations.

With structured training, appropriate governance and continued scientific oversight, these tools can help young chemists learn faster, communicate more effectively and approach problems from a broader perspective without compromising intellectual property, client confidentiality or scientific integrity. The opportunity is not to automate scientific judgement, but to equip scientists to use emerging technology with greater skill and responsibility.

Complex Synthesis That Keeps Discovery Moving: Libraries, Chiral Molecules and Advanced Modalities

Discovery programs depend on synthesis that is both creative and practical. A compelling design hypothesis has little value if the molecule cannot be made, purified, characterized and delivered in time to influence the next experiment. Complex synthesis is therefore not a support function. It is a strategic enabler of discovery velocity.

The challenge varies by program. Some teams need rapid analogue synthesis for SAR. Others need focused libraries, chiral compounds, metabolites, reference standards, peptides, carbohydrates, nucleosides, lipids, PROTACs or oligonucleotide-related chemistry. Each requires different synthetic planning and analytical support.

Custom synthesis must balance speed and quality

Fast synthesis should not compromise compound quality. Impurities, incorrect structures, unresolved stereochemistry or unstable products can mislead biology. A good synthetic partner combines route creativity with strong analytical characterization, purification capacity and communication.

Jubilant Biosys offers synthetic chemistry across Noida and Greater Noida facilities with expertise in heterocycles, carbohydrates, peptides, nucleosides, nucleotides, lipids, photo-redox chemistry, organometallics, PROTACs and oligonucleotide synthesis. Capabilities include complex molecules from grams to multikilos, focused libraries, large libraries, metabolite synthesis, scaffold synthesis, flow chemistry and multi-step stereoselective synthesis.

Libraries should answer focused questions

Parallel and library synthesis are powerful when they are designed around clear SAR questions. A focused library can explore vectors, substituent effects, physicochemical space, selectivity hypotheses or metabolic liability fixes. A large library can support broader screening or hit expansion, but only if diversity, quality and assay relevance are considered.

Synthetic teams should work with CADD, medicinal chemistry and assay biology to define the library design. The best libraries are not merely collections of compounds. They are experiments in chemical space.

Chiral and advanced chemistry need analytical discipline

Chirality can strongly influence potency, selectivity, metabolism and safety. Stereoselective synthesis, chiral resolution and chiral analytical methods are essential when stereochemistry matters. Advanced modalities such as PROTACs, lipids and oligonucleotide-related compounds add purification, characterization and stability challenges.

Analytical support is critical. LC-MS, NMR, chiral HPLC, preparative purification, HRMS and specialized methods help ensure that compounds are data-ready. This is particularly important when molecules are high molecular weight, UV-inactive, sensitive or structurally complex.

Synthesis can de-risk development early

Even in discovery, synthetic choices can influence the development path. A route that depends on scarce raw materials, hazardous conditions or chromatographic purification may be acceptable for a few milligrams but problematic later. Early awareness of route scalability helps teams avoid series that are attractive biologically but difficult to advance.

Jubilant Biosys’ integration of synthesis, medicinal chemistry, analytical chemistry, PR&D and CDMO capabilities gives sponsors a route from discovery molecules to larger-scale needs. For fast-moving programs, that continuity can reduce late hand-off risk and keep chemistry aligned with program milestones.

Talk to Jubilant Biosys about synthetic chemistry support for custom molecules, focused libraries and advanced modalities.

The Modern DMTA Cycle: Why MedChem Works Better When Biology, CADD and DMPK Share the Same Clock

The design–make–test–analyze (DMTA) cycle is the core engine of small molecule drug discovery, but it delivers value only when all components move in sync and data is interpreted collectively. When design is delayed by late DMPK insights, biology outputs lack mechanistic context, or chemistry outpaces data understanding, programs risk being active without gaining insights.

A modern DMTA cycle synchronizes medicinal chemistry, computational chemistry, assay biology, DMPK, analytics and pharmacology. The objective is not merely to turn the crank. It is to make each cycle more informative than the last.

Design should integrate multiple data streams

The design stage should begin with a shared review of the most important data. Which analogues improved potency? Which harmed permeability? Where did metabolic instability appear? Did structural biology confirm the binding pose? Did cellular potency track with biochemical potency? Did a safety signal emerge? Which compounds were hard to synthesize or purify?

When CADD, medicinal chemistry, assay biology, and DMPK evaluate these insights together, the next compound set becomes more targeted, enabling hypothesis-driven design rather than broad, unguided exploration.

Make should be fast but purposeful

Synthetic speed creates value only when aligned with the design intent. Parallel synthesis, library synthesis, scaffold and building-block access, route creativity and analytical support can accelerate compound delivery. However, chemistry teams must avoid generating large numbers of analogues that do not answer the program’s key questions.

Jubilant Biosys supports medicinal chemistry, synthetic chemistry, analytical chemistry and specialized chemistry capabilities across heterocycles, asymmetric chemistry, carbohydrates, nucleosides, nucleotides, peptides, lipids, PROTACs and oligonucleotide synthesis. This breadth helps teams choose chemistry that matches the program need.

Test and analyze should not be separated

Testing should generate interpretable, decision ready data. Potency must be assessed in context of assay conditions, compound quality, solubility, DMPK, permeability, clearance, and structural hypotheses. A less potent compound may still be valuable if it improves exposure or mitigates safety risks, while a highly potent molecule may be unsuitable if it lacks developability.

The analyze stage is critical in the DMTA cycle, where teams interpret outcomes. It should clearly identify which hypotheses hold, which fail, and define the next set of hypotheses to test, ensuring each cycle drives smarter decisions.

Shared cadence creates decision velocity

Coordinated execution across DMTA reduces cycle lag. DMPK automation and high throughput ADME can provide feedback at a cadence aligned with chemistry needs, while assay biology can integrate primary and orthogonal assays to build reliable SAR. CADD continuously refines models with emerging data, and medicinal chemistry prioritizes compounds that address key hypotheses.
When all disciplines operate on a synchronized timeline, programs learn faster—yielding not just more compounds, but a clearer path to candidates with balanced potency, pharmacokinetics, safety, and developability.

Speak with Jubilant Biosys about synchronized DMTA execution across chemistry, biology, CADD and DMPK.

Protein Quality Is Program Quality: Gene-to-Structure Strategies for Better Discovery Decisions

High-quality protein is one of the most important foundations in drug discovery. It affects assay performance, screening reliability, biophysical measurements, crystallography and structure-based design. Poor protein quality can create false negatives, false positives, irreproducible data and misleading SAR. Protein sciences therefore influence the entire discovery trajectory. 

For many targets, obtaining suitable protein is not straightforward. Construct boundaries, expression system, solubility, post-translational modifications, cofactors, complexes, purification method and storage conditions can all determine whether the protein is fit for purpose. 

Gene-to-protein strategy should match the end use 

A protein intended for a biochemical assay may need different characteristics from one intended for crystallography, SPR or fragment screening. Assay protein must be active and stable under assay conditions. SPR protein may require specific immobilization or tagging strategies. Crystallography may require construct engineering, truncations, mutations, complex formation or removal of flexible regions. 

Jubilant Biosys provides protein expression and purification services across E. coli, baculo-insect and mammalian systems, along with construct design, expression optimization, scale-up, co-expression, protein complexes, refolding, chromatography purification and protein QC. These capabilities support biochemical and biophysical assays, high-throughput screening and structural studies. 

Quality control prevents downstream ambiguity

Protein QC should not be minimal. Purity, identity, intact mass, aggregation state, activity, stability, buffer compatibility and post-translational modifications can all affect the outcome of downstream experiments. Thermal shift assays, mass spectrometry, SDS-PAGE, Western blotting, size exclusion chromatography and biophysical analysis help define whether the protein is suitable for its intended use. 

If a screen produces low hit rates or inconsistent data, the problem may be protein quality rather than chemical matter. If a structure cannot be solved, construct design or crystallization strategy may need revision. If SPR binding appears nonspecific, immobilization or protein behavior may be responsible. 

Protein sciences strengthen integrated discovery

Protein science is most valuable when connected to assay biology, CADD, medicinal chemistry and structural biology. A protein construct that supports both assay development and crystallography can accelerate the transition from hit finding to structure-based optimization. Protein QC data can explain assay variability. Structural insights can guide medicinal chemistry. SPR kinetics can add mechanistic understanding to potency values. 

At Jubilant Biosys, protein sciences and crystallography can be integrated with computational chemistry, medicinal chemistry, assay biology and DMPK. This creates a faster path from target reagent to validated hit and structure-informed lead optimization. 

Better reagents create better decisions 

In discovery, teams often focus on compounds. But the quality of the target reagent is equally important. Reliable protein enables reliable assays, reliable binding data and reliable structures. That reliability reduces wasted chemistry cycles and strengthens confidence in the program. 

A disciplined gene-to-structure strategy helps ensure that the biology the team is testing is real, reproducible and actionable. For challenging targets, that foundation can determine whether the program moves forward or stalls. 

Talk to Jubilant Biosys about protein production and structural biology support for your discovery program.