I decided that this weekend I would figure out how to create a mixture of vegetable protein powders that most closely matches that of beef. This is based on the generally held belief that the "best, most human-like" source of protein for people who perform strength training workouts is beef.
Being a non-beef eater, I decided to look at the overall amino acid profile of beef, using a flank steak as the example. Using the wonderful amino acid data available on Nutrition Data you end up with the following breakdown - the second column is the milligrams of each amino acid per 100g serving, and the third column is the normalized % composition of each amino acid:
I then retrieved the amino acid profiles of the 5 vegetable protein powders that I was planning to use in my evaluation - pea/gemma, rice, hemp, potato protein isolate, and Sacha inchi protein isolate.
To determine the optimal mixture of vegetable proteins that would most closely match the % composition of beef, I setup a non-linear solver problem in Excel. I configured the problem to solve for the minimum of the sum of the squared differences between the percent of each amino acid in the vegetable mixture vs. beef. Solving this problem yields the closest possible match.
The results were a bit surprising:
The mixture show above resulted in the following amino acid profile. The values for beef are shown for comparison:
You are welcome to download and experiement with my original Excel spreadsheet.
So my next custom protein mixture order will be something along the lines of:
60% gemma/pea protein
15% hemp protein
10% sacha inchi protein
15% potato protein
This mixture will give you all of the 9 essential amino acids, and it will deliver them in a ratio that is similar to that found in beef.
Why did I not include soy protein isolate? Primarily because of the known estrogenic effects that soy isolates are known to have.
Why did I use Excel for this? Because the solver features in Google Sheets and OpenOffice Calc only handle linear solver problems. Other commercial solver engines that hand non-linear problems often cost in the hundreds or thousands of $$.