31 October 2023

How to make predicting products of chemical reactions easier in your lab

A continued challenge for chemists worldwide has been the precise anticipation of chemical reaction outcomes. Predicting products of chemical reactions can be complex and intricate work, which often leaves a level of uncertainty that can harm the efficiency of the experiment itself. This is because chemical reactions are developed with a multitude of variables, including the reaction conditions, the properties of reactants, and the nuances of how the molecules themselves interact with each other. Calculating accurate predictions that encompass all the variables is no small task, but it is crucial for chemical development.
However, chemists no longer need to work alone in predicting the products of chemical reactions. With the help of machine learning, experiments can be designed with more confidence than ever before - helping improve lab efficiencies in time, money and resources.

Read on to learn more about how machine learning models, like SmartChemistry®, are changing the game for predicting chemical reactions.

The challenges of traditional methods

For decades, anticipating chemical reaction outcomes has formed a large and complex part of the chemist’s profession. The more traditional methods have their limitations, which can cause inefficiencies for the chemist and wider organisations. There are three key reasons why these methods can fall short in ensuring accurate predictions.


Chemical reactions are remarkably complex processes which require a wide range of variables to be monitored and controlled - from temperature and pressure to the concentration of reactants. The more variables introduced into a reaction, the harder it is to accurately predict how they will interact with each other and which products will be produced. Therefore, more complicated experiments become much more difficult to accurately predict, making them much more costly and risky for organisations to invest in.


The models and frameworks traditionally used for predicting products of chemical reactions are built on simplified assumptions. This is deliberate - the incompleteness of the models enables a level of order, a starting point and a clear trajectory for the basis of the experiment. However, these approximations can fall short in capturing the intricacies of real-world reactions. This leads to unexpected results where the experimental observations do not align with the predictions generated.

The models are built upon established chemical principles, which means that they can be particularly ineffective for chemists working with novel reactions. The undiscovered is likely to defy the predictive capabilities of the models, meaning that chemists must return to basics - generally, through trial and error.

Trial and error

The majority of traditional predictive methods rely heavily on trial and error experimentation. This requires a substantial investment of time and resources, causing inefficiencies and leaving room for human error. At a time where the pace of scientific discovery is accelerating, the inability to quickly and accurately predict reaction products can be highly detrimental to chemistry organisations.

These challenges highlight the pressing need for a more efficient and precise solution. By harnessing the computational prowess of algorithms, machine learning offers a promising avenue to overcome these longstanding hurdles and pave the way for a more efficient chemistry industry.

Machine learning and chemistry

At its core, machine learning is a subset of artificial intelligence that enables computers to learn and make predictions based on data. By feeding algorithms with copious amounts of data on  known reactions, including reactants, conditions and products, the model learns to recognise patterns, relationships and trends within the data. This learning process allows machines to make educated predictions about how similar reactions will unfold.

The transformative power of machine learning has already been adopted in a wide variety of fields, including healthcare, finance and natural language processing. It has been used for predicting diseases with unprecedented accuracy in the healthcare industry, based on patient data. However, some see a critical roadblock in bringing machine learning to chemistry. Machine learning models require complete, accurate and reliable data to train from, which many believe is missing from the chemistry field. But what if the required data was available for chemists?

One of the key differentiating features in the SmartChemistry® platform is our fundamentally different data. It is collected in real time, with both positive and negative data being recorded to ensure a complete understanding of your experiment and enabling the model to accurately predict reactions.

Benefits of using ML models in your lab

In chemistry, the advantages of machine learning are great. The technology offers a suite of invaluable advantages that can transform the way experiments are conducted and insights are gained within the laboratory. There are several specific benefits that ML brings to the lab, optimising reaction predictions for chemists.

Time efficiency

Time is a precious resource in the laboratory, but machine learning dramatically reduces the time spent on trial and error experimentation. The time taken for chemists to manually predict reaction products is cut to mere seconds by the machine learning algorithms, saving weeks or even months of laboratory time. The time savings then enable researchers to focus their efforts on more innovative and exploratory aspects of their work.

Resource savings

As we know, time is money, but that isn’t the only way ML algorithms can save on resources in the lab. Optimised experiments cost less, as the successful outcome is reached faster, and they even produce less waste to improve sustainable practices in the workplace. Fewer resources need to be purchased to complete the experiments as they are repeated less, so less money needs to be spent on these materials. Finally, optimised experiments mean less lab time for the chemists, saving significant time and ensuring that their output is greater as they can work on more projects.

Heightened accuracy

Machine learning models exhibit astonishing accuracy, capable of predicting reaction outcomes with a degree of precision that surpasses traditional methods. This enables chemists to experiment with confidence, knowing that the predicted routes are based on reliable data foundations. This is particularly important in fields like pharmaceutical research or materials science, where the slightest deviation can have profound consequences on the experiment outcomes.

Removing bias

Chemists work in a way that stems from their training in logical assumptions and understanding. However, this can sometimes be a blocker that prevents them from seeing the full picture of an experiment. Machine learning models do not have the biases that chemists have picked up, so they treat every element of data with the same level of interest and analysis. This would be impossible for chemists due to the sheer volume of data each experiment generates, but the algorithms can sort through this quickly and efficiently to spot patterns that may be undetected by the human eye.

Machine learning models serve as invaluable tools in the lab. Their predictive capabilities can save time, resources and effort for chemists while uncovering new reactions and insights that propel scientific progress.

Future directions and challenges

As machine learning continues to evolve within the chemistry field, it opens up exciting new realms of possibility. However, like any transformative technology, there are still further areas to be explored before machine learning can reach its full potential in chemistry laboratories.


As researchers continually develop new algorithms, models and techniques tailored to the needs of chemical predictions, it becomes increasingly likely that these models will offer a proliferation of specialised tools and solutions that are even more optimised for the specific needs of your lab.

Data quality and quantity

A key issue with some machine learning models is their heavy reliance on the quality and quantity of the data they are trained on. Access to comprehensive and diverse datasets is therefore crucial for the development of robust models. At deepmatter®, we are proud to be setting the standard for truly differentiated datasets. The SmartChemistry® platform is trained on highly optimised data, ensuring that the model is paving the way for high-quality machine learning in chemistry.


The successful application of machine learning in chemistry relies on a foundation of collaboration, both within the chemistry field and beyond, branching into data scientists and computer engineers. Developing effective communication and collaboration strategies is a challenge that the scientific community must address to fully harness the potential of this technology.

The SmartChemistry® platform holds collaboration in its centre, with cloud-based access enabling scientists to work collaboratively from anywhere in the world. It encompasses the very best of collaboration and confidentiality, as the cloud storage is secure and only authorised personnel is allowed access, so chemists can rest assured that their collaborative work stays in the right hands.

A transformative partnership

The marriage of chemistry and machine learning can help make predicting products of chemical reactions a more optimised, simplified and efficient process, offering a range of key benefits to chemists and helping to accelerate the discovery of groundbreaking chemical outcomes.

The era of machine learning in chemistry is here, and the power of this transformative technology is already reinventing the way chemists experiment. To find out more about SmartChemistry® and how our tools can help chemists to better design, experiment and understand their synthetic outcomes, get in touch.