Researchers improve water filter systems using AI

Sorting compounds by size and charge

The difficulty in filtering is due to the fact that synthetic porous materials are usually limited to sorting and separating compounds by either size or charge. However, biological membranes have pores made of proteins, such as aquaporin (AQP).

Aquaporins are integral membrane proteins – a protein that is permanently attached to the biological membrane – that act as channels in the transfer of water, and in some cases, small solutes across the membrane.

Aquaporin can separate water from other molecules by both size and charge because of the various distinct types of functional groups, or collections of atoms, lining the channels. M. Scott Shell, the Myers Founders Chair Professor and Vice Chair of Chemical Engineering at the University of California Santa Barbara, decided to create the same design using synthetic porous material.

Shell and his team wanted to use artificial intelligence and computers to design the inside of a carbon nanotube pore to filter boric acid-containing water.

The researchers mimicked a carbon nanotube channel with hydroxyl (water-attracting) and/or methyl (water-repelling) groups tied to each atom on the inner wall. Next, they designed and tested thousands of functional group patterns using AI, with optimization algorithms and machine learning, to assess how quickly water and boric acid would move through the pore.

Results and the future designs from the study

The study showed that artificial intelligence was useful in developing water purification membranes with new properties and could form the basis of a novel filtering system using algorithms and machine learning. Researchers came across three key findings.  

The first discovery was that optimal patterns had one or two rows of hydroxyl groups squeezed in between methyl groups, forming rings around the midsection of the pore.

The second finding was that water went through the pore nearly twice as fast as boric acid in the simulations.

The third finding showed that other neutral solutes, including phenol, benzene and isopropanol, could also become separated from water with the optimized carbon nanotube designs in a group of simulations.

The research team hopes the overall approach could be used to design and create surfaces that could have unique interactions with water or other molecules.