What do you think of when you hear the word predict? Before I started grad school, I probably would have answered that question with crystal balls and fortune tellers. In reality, scientific predictions are calculations based on empirical evidence – not gut feelings. When used appropriately, they play an integral role in expediting academic and industrial workflows, reducing instrument time, and ultimately, saving money.
As analytically minded people, we are often inclined to do each and every measurement ourselves so that we can be confident in our data. However, this approach can quickly become impractical when dealing with large sets of experiments and data. While prediction software can act as a time-saving solution to this issue, you must ensure that you are confident in its results. This is why it is important to consider the following factors when considering predictor software:
- Predictions must be performed using appropriate algorithms;
- The prediction database must contain a wide variety of empirical properties in significant quantities;
- (Optional) The prediction algorithms should have training capabilities.
The take-home message is that not all prediction software is created equally. It is wise to be selective about your software and to verify that the prediction database is a good fit for your data. If you’re working with niche and patent-protected molecules, investing in prediction software with algorithms that can be trained to work with a custom database can be an asset. Assuming that you have selected reliable prediction software, let’s explore how it can be applied.
Performing Batch Property Predictions
In the pharmaceutical industry, predictions play a key role in drug development. Prior to selecting a drug candidate, hundreds of thousands of chemicals are evaluated to determine their potency, selectivity, solubility, stability, and toxicology. In silico predictions are used to simplify this process by calculating the PhysChem, ADME, and toxicology properties of hits to further determine their viability as leads. In scenarios where molecules need to be evaluated en masse, it is significantly more practical to perform batch predictions rather than measure the properties of thousands of compounds.
Reducing or Eliminating Experiment Time and Costs
On a smaller scale, laboratories working on a tighter budget can reduce instrument costs and lost experiment time by predicting their experimental conditions. For example, method development for chromatography can be very tedious since it involves using iterative methods to investigate the multitude of parameters needed to achieve optimal separation. However, method development predictions (or simulations as they are more commonly referred to) significantly reduce the amount of time analysts spend on this process and save money spent on columns and solvents.
In certain cases where the cost of instrumentation is high and it is used infrequently, it is possible to fully substitute experiments with predictions. The most common example of this is using NMR predictions for teaching purposes. While there are low-cost NMR magnets out there, they typically sell in the 6 figure range. As an alternative, it is possible to acquire spectra from open experimental databases. However, it is time-consuming to manually sift through this data and determine which spectra are appropriate to use for teaching. Predicting NMR spectra is a cost-effective and simple way to acquire clean 1D and 2D spectra that are suitable for teaching purposes without having an instrument on-hand.
Verifying Your Chemical Structure
In chemical synthesis, we are often aware of what the end product should be, but we need structure analysis to verify whether we are correct. For NMR, predictors can be used to simulate spectra of an expected compound to assist in determining whether it is a good match to the experimental spectra. While manually comparing predicted and experimental spectra is beneficial, processing software often comes with built-in NMR predictors that will automatically verify whether your proposed structure is a good fit. As more instances where the wrong structure was passed as correct are being brought to our attention, these features are increasingly becoming essential tools.
How Does Cost Factor in?
If you’re considering incorporating predictions into your workflow, you’ll need to take their affordability into account. Purchasing prediction software is especially beneficial for large-scale endeavors where batch predictions and automated structure verification make a significant impact. However, if you’re on a budget, there are plenty of free prediction sources. These include ACD/ChemSketch freeware, which has LogP and other molecular property predictions, ACD/iLab, where users can test our PhysChem predictions, and the ACD/Spectrus Processor free trial, which has built-in NMR prediction for performing automated structure verification.
Regardless of which predictors you decide to use, it is important to ensure that they are accurate, use optimized algorithms, and effectively integrate into your workflow.
Have a question about using predictors? Contact an expert at ACD/Labs.