Collaborative research

Reactions

From mechanistic interpretation to automated discovery

Why does a reaction follow one pathway rather than another? Our collaborative studies have used electronic-structure calculations, conformational analysis and reaction-path modelling to compare intermediates, transition states and competing products.

A sustained collaboration with Amit Basak’s group examined Garratt–Braverman and Bergman cyclisations, competing rearrangements and the regioselectivity of reactions involving p-benzynes. Work with Jayanta K. Ray’s group addressed palladium-catalysed Heck cyclisations and related annulation pathways. Collaborations with Sukanta Mandal’s group investigated water-oxidation catalysts, catalyst deactivation and copper-flavonolate oxygenolysis.

The same mechanistic questions arise in very different chemical settings. Work with Pankaz K. Sharma and Parayil Kumaran Ajikumar examined the cytochrome P450-catalysed formation of the oxetane ring during Taxol biosynthesis. Theoretical collaboration with Biswarup Pathak explored oxygen reduction on platinum subnanoclusters, where structure, electronic state and catalytic behaviour must be considered together.

These studies showed both the power and the limitation of conventional mechanistic modelling. Once plausible intermediates and products have been proposed, calculations can distinguish among them. But the search remains constrained by what someone thought to propose; unexpected products and alternative pathways may never enter the calculation.

This difficulty led us towards automated reaction discovery: beginning with reactants, systematically generating possible encounters and bond rearrangements, identifying distinct products and then examining the pathways that connect them. It also contributed directly to the development of PyAR as a framework for exploring reaction space rather than calculating only a mechanism chosen in advance.

A collaboration with Mario Barbatti provided another decisive turn. Our study of the photochemical transformation of an HCN tetramer combined excited-state dynamics, ground-state pathways and kinetic modelling to examine a step towards a purine precursor. It showed that the source and dissipation of energy can change which mechanisms are accessible: a reaction map restricted to thermal ground-state chemistry may miss the relevant pathway.

That work introduced prebiotic chemistry into our programme. The questions it raised—how simple molecules generate unexpectedly diverse products, and how thermal, photochemical and environmental conditions reshape the available pathways—subsequently became part of our research on automated reaction discovery and prebiotic chemical space.

Structures

From interpreting a geometry to discovering structural landscapes

Structures reported experimentally often lead to deeper questions. Why is one geometry preferred? Is the bonding adequately represented by a conventional Lewis structure? Could other stable arrangements exist but remain unobserved?

Our collaborations have examined such questions in molecular complexes, metal clusters, non-covalently bound aggregates, nanoalloys and systems with unusual coordination or ligand hapticity. Theoretical work with P. K. Chattaraj explored all-metal and all-pnictogen aromaticity, planar tetracoordinate atoms and nonclassical bonding in clusters and nanoalloys. Collaboration with Venkatesan S. Thimmakondu investigated planar hypercoordination, organomagnesium systems and other chemically unconventional regions of molecular space.

These studies made another limitation clear. A proposed structure can be optimised and characterised with considerable precision, but that calculation cannot determine whether a more stable structure was never considered. The problem becomes especially severe for clusters and aggregates, where the number of possible geometries grows rapidly with size and composition.

This prompted us to move from structure interpretation to structure discovery. In PyAR, structures can be grown recursively by bringing atoms, molecules or fragments together in different orientations, optimising the resulting candidates and removing equivalent structures. The same general strategy can be applied across molecular aggregates, metallic clusters and multicomponent nanoalloys.

What began as an effort to explain unusual experimental and theoretical structures therefore developed into a broader programme on the automated exploration of structural landscapes.

Properties

From explaining observations to designing function

A third group of collaborations asks how molecular structure and environment determine measurable behaviour. Here we calculate excited states, spectra, energetics, binding interactions and molecular dynamics, and compare them with experimental observations.

Our long-standing collaboration with N. D. Pradeep Singh’s group has addressed photoacid generators, photoremovable protecting groups, excited-state proton transfer, fluorescent monitoring, phototherapeutics and light-induced biological activity. These projects required us to connect electronic structure and excited-state processes with experimentally observed photochemical behaviour.

Related collaborations include photoluminescence in silver clusters with Di Sun’s group, thermally activated delayed-fluorescence emitters and OLEDs with Joshy Joseph and colleagues at CSIR–NIIST, and biomolecular binding and stability with Biju A. R. Work with Subhas Ghosal connects calculated energies and spectroscopic signatures with the possible identification of molecules in astrochemical environments.

Repeatedly calculating why a known molecule displays a particular property suggested the inverse question: can chemical space be searched for molecules expected to display a desired property?

This shift—from interpreting properties to using them as selection criteria—led towards our work on molecular discovery and functional design. Candidate structures can be generated, screened computationally and ranked using quantities related to emission, excited-state behaviour, stability, energy content or another intended function. Current examples include TADF emitters and computationally generated high-energy-density molecular candidates.

Collaborative property studies thus became a foundation for moving from the explanation of known molecules towards the design and experimental evaluation of new candidates.

Scale and complexity

Why AI and machine learning entered the programme

Reaction networks, structural landscapes and molecular libraries all present the same practical problem: the number of candidates can become much larger than the number that can be examined with demanding electronic-structure calculations.

AI and machine learning entered our programme as a response to this scale—not as a substitute for chemical theory. Their useful roles include recognising related structures, learning structure–property relationships, ranking candidates, guiding searches and identifying where more accurate calculations would be most informative.

The objective is to combine automated exploration, validated quantum-chemical calculations and interpretable machine-learning models. Scientific software is essential to this effort because searches must remain reproducible, inspectable and connected to the chemical assumptions on which they depend.

Our present research themes were therefore not chosen as separate topics. They emerged from a continuing effort to understand experiments more completely: mechanistic studies led to automated reaction discovery; structural problems led to global structure exploration; property calculations led to functional molecular design; and the scale of all three led us towards AI-assisted molecular science.

Collaboration remains where these methods are tested against difficult chemistry—and where new research directions often begin.