Techniques for modelling and optimizing dialysis toxin displacer compounds

An apparatus, comprising:

• at least one processor;
• a memory coupled to the at least one processor, the memory comprising instructions that, when executed by the at least one processor, cause the at least one processor to:
• access candidate data comprising data of a set of candidate compounds with unknown binding affinity to albumin, and
• determine at least one potential compound with an affinity for binding to albumin from the candidate data by providing the candidate data as input to at least one machine learning quantitative structure-activity relationship (QSAR) model trained to model binding to an albumin target protein via:
• building the at least one QSAR model using first known albumin binding data comprising data of a first set of compounds known to bind to albumin;
• and
• validating the at least one QSAR model using second known albumin binding data comprising data of a second set of compounds known to bind to albumin, the second set of compounds different from the first set of compounds;
• wherein the at least one QSAR model is trained to receive the candidate data as input, the candidate data different from the first known albumin binding data and the second known albumin binding data, to generate output comprising at least one potential compound with an affinity for binding to albumin, the at least one potential compound selected based on the at least one QSAR model from the set of candidate compounds;
• generate an infusion profile for a patient based on the at least one potential compound;
• and
• cause an infusion pump of a dialysis machine to be controlled based on the infusion profile for the patient during a dialysis treatment of the patient.

A method comprising:

• training at least one machine learning quantitative structure-activity relationship (QSAR) model for modeling binding to an albumin target protein via:
• building the at least one QSAR model using first known albumin binding data comprising data of a first set of compounds known to bind to albumin;
• and
• validating the at least one QSAR model using second known albumin binding data comprising data of a second set of compounds known to bind to albumin, the second set of compounds different from the first set of compounds;
• wherein the at least one QSAR model is trained to receive candidate data as input, the candidate data comprising data of a set of candidate compounds with unknown binding affinity to albumin, the candidate data different from the first known albumin binding data and the second known albumin binding data, to generate output comprising at least one potential compound with an affinity for binding to albumin, wherein the at least one potential compound is selected based on the at least QSAR model from the set of candidate compounds;
• and
• administering a dialysis treatment using the at least one potential compound.