| Abstract: | Five laboratories collaborated in a two-part study to improve and measure the accuracy and precision of amino acid analysis. The results of the unpublished first part allowed each laboratory the opportunity to optimise their own procedure. In the second part of the study reported here, three purified and six conventional proteins were analysed, each after three different preparation procedures (normal and extended acid hydrolysis, and following preliminary oxidation). Preparative effects reported in the literature were confirmed though not uniformly, and there were other, often unique, preparative effects. Well documented preparative effects are essential in matching analytical work to differing requirements. Results from one laboratory were excluded from further statistical analysis because of high replicate variation. The repeatability (within laboratory) coefficient of variation was 4.1% and for reproducibility (between laboratory) it was 13.1%. Discrepancies between best amino acid estimates for purified proteins and estimates calculated from published sequences (P < 0.01) provided evidence of bias in five amino acids from one laboratory and in one from another. When P < 0.01 was used as the discard threshold for conventional protein results, a total of 9.9% of best estimates were discarded: 8.1% were biased, 3.2% were outliers and 1.3% were both biased and outliers. When these values were discarded, 19 of 96 mean values changed by up to 4%, nine changed by 4-8%, eight changed by larger amounts and the pooled coefficient of variation fell from 8.2% to 5.4%. At this threshold between zero and 31% of each laboratory's data on conventional proteins was discarded. A lower threshold (P < 0.05) doubled the discard rate but had marginal effects on mean values and residual variation. At P < 0.01 less than 20% of biased estimates were also outliers and less than half the outliers were also biased. This means that only studies which examine both accuracy and precision will achieve maximum improvement in either. |
