Sprecher
Beschreibung
Understanding membrane fitness for the Origins of Life Using Automation and Machine Learning
The interplay between environment and protocells is assumed to be a driving force for chemical and early Darwinian evolution. However, the number of environmental and biological variables makes predicting protocell stability and functionality an unresolved challenge in understanding environment habitability for astrobiology. To address this challenge, we have applied lab automation combined with artificial intelligence to formulation chemistry for the origins of life. Specifically, we have built a foundation model (FM) to predict liposome formation from novel amphiphiles. The model relies on latent space vectors generated from SMILES representations of chemical structures coupled to extant data sets reflecting chemical properties, here as membrane formation. Our foundation model improved liposome formation prediction from a previous Gaussian process classifier model from 0.85 to 0.90 for our best latent space descriptors.
We also explored how environments impacted the function of the membranes by determining the CVC, liposome formation, and permeability. We tested these metrics under four pH conditions, with and without sea salt or sea salt mixed with organics. We also tested how changing the headgroups of some of our amphiphiles from phosphate to sulfate impacted membrane function. We found that permeability and CVC did not follow intuitive trends. Too, while environment influenced the metrics tested, we did not eliminate any environments for poor liposome formation or find an optimal environment for the liposome compositions tested. We found that dodecyl phosphate reduced both permeability and number of liposomes formed when compared to dodecyl sulfate, highlighting the importance of phosphate headgroups form membrane formation and function.