A systematic analysis of regression models for protein engineering.

To optimize proteins for particular traits holds great promise for industrial and pharmaceutical purposes. Machine Learning is increasingly applied in this field to predict properties of proteins, thereby guiding the experimental optimization process. A natural question is: How much progress are we making with such predictions, and how important is the choice of regressor and representation? In this paper, we demonstrate that different assessment criteria for regressor performance can lead to dramatically different conclusions, depending on the choice of metric, and how one defines generalization. We highlight the fundamental issues of sample bias in typical regression scenarios and how this can lead to misleading conclusions about regressor performance. Finally, we make the case for the importance of calibrated uncertainty in this domain.

Change in the intrathyroidal kinetics of radioiodine under continued and discontinued antithyroid medication in Graves' disease.

PURPOSE: This study evaluated the thyroidal kinetics of radioiodine in Graves' disease under continued thiamazole medication and after discontinuation of thiamazole for 1-2 days, with a view to keeping the period of discontinuation as short as possible and to exploring the underlying mechanism of a postulated radioprotective effect of antithyroid drugs. METHODS: In 316 patients, diagnostic and therapeutic radioiodine kinetics were followed up for 2 days by ten uptake measurements each and were defined mathematically by a two-compartment model. RESULTS: Without thiamazole or when thiamazole was discontinued for at least 2 days, all uptake curves could be fitted perfectly by a simple in- and output function; the mean square error (mse) was 0.38 (test) and 0.28 (therapy). Under continued thiamazole medication (11.0+/-7.0 mg/day), the energy dose delivered to the thyroid was lowered by factor of 2.5. Uptake curves were deformed (mse: 1.06, test and 0.86, therapy) and appeared two peaked, suggesting coexistence of follicles with blocked and follicles with intact hormone synthesis and hence heterogeneous radioiodine uptake in the thyroid. In patients with maximally altered uptake curves, the success rate was as low as 31%. One day after discontinuation of thiamazole, mse was still increased (0.78, test), while 2 days afterwards it had normalised (0.36, test) and 3 days afterwards (mse: 0.24, therapy) the success rate was 87%. CONCLUSION: Efficacy of radioiodine therapy under continued thiamazole medication is reduced not only by a lower uptake and shorter half-life of radioiodine, but also by a heterogeneous energy dose distribution in the thyroid. Discontinuation of thiamazole (but probably not of propylthiouracil) for at least 2 days is required to restore the efficacy of radioiodine.

Incidence of radiation-induced Graves' disease in patients treated with radioiodine for thyroid autonomy before and after introduction of a high-sensitivity TSH receptor antibody assay.

Autoimmune hyperthyroidism may occur several months after radioiodine therapy (RIT) for functional thyroid autonomy. Exacerbation of pre-existing subclinical Graves' disease (GD) has been held responsible for this phenomenon. Determination of TSH receptor antibody using solubilised porcine epithelial cell membranes is insensitive and may have failed to diagnose GD in these patients before RIT. Following the introduction of a more sensitive assay, using the human TSH receptor as an antigen, it has been expected that the incidence of radiation-induced GD after RIT for functional thyroid autonomy will be reduced. In a first group of 1,428 patients treated between November 1993 and March 1997 (group I) we used the porcine TRAb assay to exclude GD, while in a second group comprising 1,408 patients treated between January 2000 and December 2001 (group II), GD was excluded using the human TRAb assay. A matched control group of 231 patients was derived from group II. In group I a total of 15 (1.05%) patients developed obvious or suspected radiation-induced GD, while in group II 17 (1.2%) did so; the interval until development of GD was 8.4 and 8.8 months, respectively, after RIT. Serum anti-thyroid peroxidase levels before RIT were elevated in 36.4% of group I patients and 47.1% of group II patients, but in only 5.6% of the control group. Other non-specific signs of mild immunopathy of the thyroid were seen retrospectively in 73.3%, 64.7% and 16.0% of the patients in these three groups, respectively. In conclusion, the introduction of a high-sensitivity TRAb assay did not reduce the incidence of autoimmune hyperthyroidism occurring late after RIT for functional thyroid autonomy, but mild immunopathy of the thyroid is seen more frequently in these patients and seems to be a predisposing factor in the development of radiation-induced GD.