PHARMACEUTICALS
Table 1- Key reaction types in oncology API synthesis & what continuous flow resolves
like LipoHedge further extend the clinical value of AI-identified molecules. Liposomes protect drugs from degradation, improve cellular uptake, and reduce off-target toxicity. LipoHedge formulations incorporate targeted delivery mechanisms and enhanced stability.
Together, these technologies achieve lower, more tolerable doses, reduced systemic toxicity, extended circulating half-life, and passive tumour targeting via the EPR( enhanced permeability and retention) effect. This synergy between AIdesigned molecules, continuous flowsynthesised actives and advanced formulations is where competitive advantage now resides.
Market drivers
The commercial case follows the chemistry. Cost is driven by step count, cumulative yield loss, precious metal catalysts, containment infrastructure and regulatory compliance. Removing steps lowers the ever-larger cost lever. The HPAPI market is forecast to grow from approximately $ 25 billion in 2024 to nearly $ 40 billion by 2029. Small molecules continue to account for the majority of novel drug approvals. This is not niche demand.
In January 2025, the US FDA issued its first draft guidance on AI in drug development, proposing a risk-based credibility framework. 4 Notably, it
excludes discovery itself, where a wrong prediction merely results in waste of the experiment process. The regulatory focus falls on nonclinical, clinical, post-marketing and manufacturing applications.
Case study: AI compound analysis
Consider a compound identified through AI screening: SMILES notation O = n1c( nc( cc1NC(= O) C) N1CCC1) NC. Molecular weight calculation gives 279.3 g / mol, with a logP of 0.23( balanced solubility). The topological polar surface area( TPSA) was 101.27 Ų, supporting water solubility and oral bioavailability. All of these parameters satisfy Lipinski’ s rule of five, indicating druglike properties.
ML models predict random-forest classifiers to score 62.5 % anti-cancer probability, gradient-boosting models reach 99.1 % and neural networks score 92.7 %. The ensemble verdict is‘ likely anti-cancer’, with 75 % agreement across models. The next steps are definitive: in vitro cytotoxicity assays on cancer cell lines, in vivo pharmacokinetic studies in animal models and computational modelling to optimise structure.
Outlook
Value from these developments will accrue not to general CDMOs, but to organisations that invest in data infrastructure before models, keep experienced synthetic chemists in the loop, and connect powerful digital tools to supply chains that can actually deliver the compound. Continuous flow chemistry, advanced formulations, and AI-integrated discovery can help to solve the synthesis challenge.
The bottom line is that the molecule still has to be made and this is precisely where chemistry expertise, process innovation and manufacturing capability are required to compete in drug discovery. ●
References: 1: Z. Xu, F. Ren, P. Wang et al., Nature Medicine, 2025, 31( 8), 2602 – 2610. 2: Pharmaceuticals( Basel), 2026, 19( 6), 916. 3: J. Abramson, J. Adler, J. Dunger et al., Nature, 2024, 630( 8016), 493 – 500. 4: US FDA, Draft Guidance, FDA-2024-D-4689, January 2025.
J j
Subir Chakraborty
CEO
CD CHEM GROUP subir @ cdchemgroup. com www. cdchemgroup. com
SEP / OCT 2026 SPECCHEMONLINE. COM
35