Bayesian nonparametric latent variable models

L’un des problèmes importants en apprentissage automatique est de déterminer la complexité du modèle à apprendre. Une trop grande complexité mène au surapprentissage, ce qui correspond à trouver des structures qui n’existent pas réellement dans les données, tandis qu’une trop faible complexité mène au sous-apprentissage, c’est-à-dire que l’expressivité du modèle est insuffisante pour capturer l’ensemble des structures présentes dans les données. Pour certains modèles probabilistes, la complexité du modèle se traduit par l’introduction d’une ou plusieurs variables cachées dont le rôle est d’expliquer le processus génératif des données. Il existe diverses approches permettant d’identifier le nombre approprié de variables cachées d’un modèle. Cette thèse s’intéresse aux méthodes Bayésiennes nonparamétriques permettant de déterminer le nombre de variables cachées à utiliser ainsi que leur dimensionnalité. La popularisation des statistiques Bayésiennes nonparamétriques au sein de la communauté de l’apprentissage automatique est assez récente. Leur principal attrait vient du fait qu’elles offrent des modèles hautement flexibles et dont la complexité s’ajuste proportionnellement à la quantité de données disponibles. Au cours des dernières années, la recherche sur les méthodes d’apprentissage Bayésiennes nonparamétriques a porté sur trois aspects principaux : la construction de nouveaux modèles, le développement d’algorithmes d’inférence et les applications. Cette thèse présente nos contributions à ces trois sujets de recherches dans le contexte d’apprentissage de modèles à variables cachées. Dans un premier temps, nous introduisons le Pitman-Yor process mixture of Gaussians, un modèle permettant l’apprentissage de mélanges infinis de Gaussiennes. Nous présentons aussi un algorithme d’inférence permettant de découvrir les composantes cachées du modèle que nous évaluons sur deux applications concrètes de robotique. Nos résultats démontrent que l’approche proposée surpasse en performance et en flexibilité les approches classiques d’apprentissage. Dans un deuxième temps, nous proposons l’extended cascading Indian buffet process, un modèle servant de distribution de probabilité a priori sur l’espace des graphes dirigés acycliques. Dans le contexte de réseaux Bayésien, ce prior permet d’identifier à la fois la présence de variables cachées et la structure du réseau parmi celles-ci. Un algorithme d’inférence Monte Carlo par chaîne de Markov est utilisé pour l’évaluation sur des problèmes d’identification de structures et d’estimation de densités. Dans un dernier temps, nous proposons le Indian chefs process, un modèle plus général que l’extended cascading Indian buffet process servant à l’apprentissage de graphes et d’ordres. L’avantage du nouveau modèle est qu’il admet les connections entres les variables observables et qu’il prend en compte l’ordre des variables. Nous présentons un algorithme d’inférence Monte Carlo par chaîne de Markov avec saut réversible permettant l’apprentissage conjoint de graphes et d’ordres. L’évaluation est faite sur des problèmes d’estimations de densité et de test d’indépendance. Ce modèle est le premier modèle Bayésien nonparamétrique permettant d’apprendre des réseaux Bayésiens disposant d’une structure complètement arbitraire.

'Bayesian Optimisation for Nurse Scheduling'

A Bayesian optimisation algorithm for a nurse scheduling problem is presented, which involves choosing a suitable scheduling rule from a set for each nurse's assignment. When a human scheduler works, he normally builds a schedule systematically following a set of rules. After much practice, the scheduler gradually masters the knowledge of which solution parts go well with others. He can identify good parts and is aware of the solution quality even if the scheduling process is not yet completed, thus having the ability to finish a schedule by using flexible, rather than fixed, rules. In this paper, we design a more human-like scheduling algorithm, by using a Bayesian optimisation algorithm to implement explicit learning from past solutions. A nurse scheduling problem from a UK hospital is used for testing. Unlike our previous work that used Genetic Algorithms to implement implicit learning [1], the learning in the proposed algorithm is explicit, i.e. we identify and mix building blocks directly. The Bayesian optimisation algorithm is applied to implement such explicit learning by building a Bayesian network of the joint distribution of solutions. The conditional probability of each variable in the network is computed according to an initial set of promising solutions. Subsequently, each new instance for each variable is generated by using the corresponding conditional probabilities, until all variables have been generated, i.e. in our case, new rule strings have been obtained. Sets of rule strings are generated in this way, some of which will replace previous strings based on fitness. If stopping conditions are not met, the conditional probabilities for all nodes in the Bayesian network are updated again using the current set of promising rule strings. For clarity, consider the following toy example of scheduling five nurses with two rules (1: random allocation, 2: allocate nurse to low-cost shifts). In the beginning of the search, the probabilities of choosing rule 1 or 2 for each nurse is equal, i.e. 50%. After a few iterations, due to the selection pressure and reinforcement learning, we experience two solution pathways: Because pure low-cost or random allocation produces low quality solutions, either rule 1 is used for the first 2-3 nurses and rule 2 on remainder or vice versa. In essence, Bayesian network learns 'use rule 2 after 2-3x using rule 1' or vice versa. It should be noted that for our and most other scheduling problems, the structure of the network model is known and all variables are fully observed. In this case, the goal of learning is to find the rule values that maximize the likelihood of the training data. Thus, learning can amount to 'counting' in the case of multinomial distributions. For our problem, we use our rules: Random, Cheapest Cost, Best Cover and Balance of Cost and Cover. In more detail, the steps of our Bayesian optimisation algorithm for nurse scheduling are: 1. Set t = 0, and generate an initial population P(0) at random; 2. Use roulette-wheel selection to choose a set of promising rule strings S(t) from P(t); 3. Compute conditional probabilities of each node according to this set of promising solutions; 4. Assign each nurse using roulette-wheel selection based on the rules' conditional probabilities. A set of new rule strings O(t) will be generated in this way; 5. Create a new population P(t+1) by replacing some rule strings from P(t) with O(t), and set t = t+1; 6. If the termination conditions are not met (we use 2000 generations), go to step 2. Computational results from 52 real data instances demonstrate the success of this approach. They also suggest that the learning mechanism in the proposed approach might be suitable for other scheduling problems. Another direction for further research is to see if there is a good constructing sequence for individual data instances, given a fixed nurse scheduling order. If so, the good patterns could be recognized and then extracted as new domain knowledge. Thus, by using this extracted knowledge, we can assign specific rules to the corresponding nurses beforehand, and only schedule the remaining nurses with all available rules, making it possible to reduce the solution space. Acknowledgements The work was funded by the UK Government's major funding agency, Engineering and Physical Sciences Research Council (EPSRC), under grand GR/R92899/01. References [1] Aickelin U, "An Indirect Genetic Algorithm for Set Covering Problems", Journal of the Operational Research Society, 53(10): 1118-1126,

'Bayesian Optimisation for Nurse Scheduling'

A Bayesian optimisation algorithm for a nurse scheduling problem is presented, which involves choosing a suitable scheduling rule from a set for each nurse's assignment. When a human scheduler works, he normally builds a schedule systematically following a set of rules. After much practice, the scheduler gradually masters the knowledge of which solution parts go well with others. He can identify good parts and is aware of the solution quality even if the scheduling process is not yet completed, thus having the ability to finish a schedule by using flexible, rather than fixed, rules. In this paper, we design a more human-like scheduling algorithm, by using a Bayesian optimisation algorithm to implement explicit learning from past solutions. A nurse scheduling problem from a UK hospital is used for testing. Unlike our previous work that used Genetic Algorithms to implement implicit learning [1], the learning in the proposed algorithm is explicit, i.e. we identify and mix building blocks directly. The Bayesian optimisation algorithm is applied to implement such explicit learning by building a Bayesian network of the joint distribution of solutions. The conditional probability of each variable in the network is computed according to an initial set of promising solutions. Subsequently, each new instance for each variable is generated by using the corresponding conditional probabilities, until all variables have been generated, i.e. in our case, new rule strings have been obtained. Sets of rule strings are generated in this way, some of which will replace previous strings based on fitness. If stopping conditions are not met, the conditional probabilities for all nodes in the Bayesian network are updated again using the current set of promising rule strings. For clarity, consider the following toy example of scheduling five nurses with two rules (1: random allocation, 2: allocate nurse to low-cost shifts). In the beginning of the search, the probabilities of choosing rule 1 or 2 for each nurse is equal, i.e. 50%. After a few iterations, due to the selection pressure and reinforcement learning, we experience two solution pathways: Because pure low-cost or random allocation produces low quality solutions, either rule 1 is used for the first 2-3 nurses and rule 2 on remainder or vice versa. In essence, Bayesian network learns 'use rule 2 after 2-3x using rule 1' or vice versa. It should be noted that for our and most other scheduling problems, the structure of the network model is known and all variables are fully observed. In this case, the goal of learning is to find the rule values that maximize the likelihood of the training data. Thus, learning can amount to 'counting' in the case of multinomial distributions. For our problem, we use our rules: Random, Cheapest Cost, Best Cover and Balance of Cost and Cover. In more detail, the steps of our Bayesian optimisation algorithm for nurse scheduling are: 1. Set t = 0, and generate an initial population P(0) at random; 2. Use roulette-wheel selection to choose a set of promising rule strings S(t) from P(t); 3. Compute conditional probabilities of each node according to this set of promising solutions; 4. Assign each nurse using roulette-wheel selection based on the rules' conditional probabilities. A set of new rule strings O(t) will be generated in this way; 5. Create a new population P(t+1) by replacing some rule strings from P(t) with O(t), and set t = t+1; 6. If the termination conditions are not met (we use 2000 generations), go to step 2. Computational results from 52 real data instances demonstrate the success of this approach. They also suggest that the learning mechanism in the proposed approach might be suitable for other scheduling problems. Another direction for further research is to see if there is a good constructing sequence for individual data instances, given a fixed nurse scheduling order. If so, the good patterns could be recognized and then extracted as new domain knowledge. Thus, by using this extracted knowledge, we can assign specific rules to the corresponding nurses beforehand, and only schedule the remaining nurses with all available rules, making it possible to reduce the solution space. Acknowledgements The work was funded by the UK Government's major funding agency, Engineering and Physical Sciences Research Council (EPSRC), under grand GR/R92899/01. References [1] Aickelin U, "An Indirect Genetic Algorithm for Set Covering Problems", Journal of the Operational Research Society, 53(10): 1118-1126,

Bayesian Estimation of Small Proportions Using Binomial Group Test

Group testing has long been considered as a safe and sensible relative to one-at-a-time testing in applications where the prevalence rate p is small. In this thesis, we applied Bayes approach to estimate p using Beta-type prior distribution. First, we showed two Bayes estimators of p from prior on p derived from two different loss functions. Second, we presented two more Bayes estimators of p from prior on π according to two loss functions. We also displayed credible and HPD interval for p. In addition, we did intensive numerical studies. All results showed that the Bayes estimator was preferred over the usual maximum likelihood estimator (MLE) for small p. We also presented the optimal β for different p, m, and k.

Sparse hierarchical Bayesian models for detecting relevant antigenic sites in virus evolution

Understanding how virus strains offer protection against closely related emerging strains is vital for creating effective vaccines. For many viruses, including Foot-and-Mouth Disease Virus (FMDV) and the Influenza virus where multiple serotypes often co-circulate, in vitro testing of large numbers of vaccines can be infeasible. Therefore the development of an in silico predictor of cross-protection between strains is important to help optimise vaccine choice. Vaccines will offer cross-protection against closely related strains, but not against those that are antigenically distinct. To be able to predict cross-protection we must understand the antigenic variability within a virus serotype, distinct lineages of a virus, and identify the antigenic residues and evolutionary changes that cause the variability. In this thesis we present a family of sparse hierarchical Bayesian models for detecting relevant antigenic sites in virus evolution (SABRE), as well as an extended version of the method, the extended SABRE (eSABRE) method, which better takes into account the data collection process. The SABRE methods are a family of sparse Bayesian hierarchical models that use spike and slab priors to identify sites in the viral protein which are important for the neutralisation of the virus. In this thesis we demonstrate how the SABRE methods can be used to identify antigenic residues within different serotypes and show how the SABRE method outperforms established methods, mixed-effects models based on forward variable selection or l1 regularisation, on both synthetic and viral datasets. In addition we also test a number of different versions of the SABRE method, compare conjugate and semi-conjugate prior specifications and an alternative to the spike and slab prior; the binary mask model. We also propose novel proposal mechanisms for the Markov chain Monte Carlo (MCMC) simulations, which improve mixing and convergence over that of the established component-wise Gibbs sampler. The SABRE method is then applied to datasets from FMDV and the Influenza virus in order to identify a number of known antigenic residue and to provide hypotheses of other potentially antigenic residues. We also demonstrate how the SABRE methods can be used to create accurate predictions of the important evolutionary changes of the FMDV serotypes. In this thesis we provide an extended version of the SABRE method, the eSABRE method, based on a latent variable model. The eSABRE method takes further into account the structure of the datasets for FMDV and the Influenza virus through the latent variable model and gives an improvement in the modelling of the error. We show how the eSABRE method outperforms the SABRE methods in simulation studies and propose a new information criterion for selecting the random effects factors that should be included in the eSABRE method; block integrated Widely Applicable Information Criterion (biWAIC). We demonstrate how biWAIC performs equally to two other methods for selecting the random effects factors and combine it with the eSABRE method to apply it to two large Influenza datasets. Inference in these large datasets is computationally infeasible with the SABRE methods, but as a result of the improved structure of the likelihood, we are able to show how the eSABRE method offers a computational improvement, leading it to be used on these datasets. The results of the eSABRE method show that we can use the method in a fully automatic manner to identify a large number of antigenic residues on a variety of the antigenic sites of two Influenza serotypes, as well as making predictions of a number of nearby sites that may also be antigenic and are worthy of further experiment investigation.

Defensive role of allelopathic secondary compounds in plants I: testing two independent general hypotheses

This study tests two general and independent hypotheses with the basic assumption that phytoactive secondary compounds produced by plants evolved primarily as plant defences against competitor plant species. The first hypothesis is that the production and main way of release of phytoactive compounds reflect an adaptive response to climatic conditions. Thus, higher phytoactivity by volatile compounds prevails in plants of hot, dry environments, whereas higher phytoactivity by water-soluble compounds is preponderant in plants from wetter environments. The second hypothesis is that synergy between plant phytoactive compounds is widespread, due to the resulting higher energy efficiency and economy of resources. The first hypothesis was tested on germination and early growth of cucumber treated with either water extracts or volatiles from leaves or vegetative shoot tops of four Mediterranean-type shrubs. The second hypothesis was tested on germination of subterranean clover treated with either water extracts of leaves or vegetative shoot tops of one tree and of three Mediterranean-type shrubs or with each of the three fractions obtained from water extracts. Our data do not support either hypotheses. We found no evidence for higher phytoactivity in volatile compounds released by plants that thrive in hot, dry Mediterranean-type environments. We also found no evidence for the predominance of synergy among the constituents of fractions. To the contrary, we found either antagonism or no interaction of effects among allelopathic compounds.

Análise bayesiana do modelo de herança monogênica no melhoramento vegetal: um exemplo com abobrinha

A common breeding strategy is to carry out basic studies to investigate the hypothesis of a single gene controlling the trait (major gene) with or without polygenes of minor effect. In this study we used Bayesian inference to fit genetic additive-dominance models of inheritance to plant breeding experiments with multiple generations. Normal densities with different means, according to the major gene genotype, were considered in a linear model in which the design matrix of the genetic effects had unknown coefficients (which were estimated in individual basis). An actual data set from an inheritance study of partenocarpy in zucchini (Cucurbita pepo L.) was used for illustration. Model fitting included posterior probabilities for all individual genotypes. Analysis agrees with results in the literature but this approach was far more efficient than previous alternatives assuming that design matrix was known for the generations. Partenocarpy in zucchini is controlled by a major gene with important additive effect and partial dominance.

Micromorphological and hardness analyses of human and bovine sclerotic dentin: a comparative study

The purpose of this study was to test the hypothesis that both human and bovine sclerotic dentin have similar hardness properties, in addition to similar micromorphological characteristics. Sixteen teeth (8 human and 8 bovine) exhibiting exposed dentin in the incisal edge and showing characteristics typical of sclerosis were used. Vickers surface microhardness testing was conducted. Three areas of the dentin surface of each specimen were selected. All teeth were processed for scanning electron microscopy in order to estimate the amount (in percentage) of solid dentin on the sclerotic dentin surface. The data were compared by Student's t test (α = 0.05). The micromorphological and microhardness data were compared by Pearson's linear correlation test (α = 0.05). The mean percentages of solid dentin of human and bovine sclerotic dentin were similar (human 90.71 ± 0.83 and bovine 89.08 ± 0.81, p = 0.18). The mean microhardness value (VHN) of human sclerotic dentin was significantly higher than that of bovine sclerotic dentin (human 45.26 ± 2.92 and bovine 29.93 ± 3.83, p = 0.006). No correlation was found between the microhardness values and the amount of solid dentin in the sclerotic dentin, irrespective of the species considered (human R² = 0.0240, p = 0.714; bovine R² = 0.0017, p = 0.923; and combined R² = 0.038, p = 0.46). We concluded that although both bovine and human sclerotic dentin present a similar amount of solid tissue, human sclerotic dentin presents higher microhardness than bovine sclerotic dentin.

Flexural and diametral tensile strength of composite resins

This study evaluated the flexural strength (sf) and the diametral tensile strength (st) of light-cured composite resins, testing the hypothesis that there is a positive relation between these properties. Twenty specimens were fabricated for each material (Filtek Z250- 3M-Espe; AM- Amelogen, Ultradent; VE- Vit-l-escence, Ultradent; EX- Esthet-X, Dentsply/Caulk), following ISO 4049 and ANSI/ADA 27 specifications and the manufacturers’ instructions. For the st test, cylindrical shaped (4 mm x 6 mm) specimens (n = 10) were placed with their long axes perpendicular to the applied compressive load at a crosshead speed of 1.0 mm/min. The sf was measured using the 3-point bending test, in which bar shaped specimens (n = 10) were tested at a crosshead speed of 0.5 mm/min. Both tests were performed in a universal testing machine (EMIC 2000) recording the fracture load (N). Strength values (MPa) were calculated and statistically analyzed by ANOVA and Tukey (a = 0.05). The mean and standard deviation values (MPa) were Z250-45.06 ± 5.7; AM-35.61 ± 5.4; VE-34.45 ± 7.8; and EX-42.87 ± 6.6 for st; and Z250-126.52 ± 3.3; AM-87.75 ± 3.8; VE-104.66 ± 4.4; and EX-119.48 ± 2.1 for sf. EX and Z250 showed higher st and sf values than the other materials evaluated (p < 0.05), which followed a decreasing trend of mean values. The results confirmed the study hypothesis, showing a positive relation between the material properties examined.

Array-CGH testing in spontaneous abortions with normal karyotypes

In about 50% of first trimester spontaneous abortion the cause remains undetermined after standard cytogenetic investigation. We evaluated the usefulness of array-CGH in diagnosing chromosome abnormalities in products of conception from first trimester spontaneous abortions. Cell culture was carried out in short- and long-term cultures of 54 specimens and cytogenetic analysis was successful in 49 of them. Cytogenetic abnormalities (numerical and structural) were detected in 22 (44.89%) specimens. Subsequent, array-CGH based on large insert clones spaced at ~1 Mb intervals over the whole genome was used in 17 cases with normal G-banding karyotype. This revealed chromosome aneuplodies in three additional cases, giving a final total of 51% cases in which an abnormal karyotype was detected. In keeping with other recently published works, this study shows that array-CGH detects abnormalities in a further ~10% of spontaneous abortion specimens considered to be normal using standard cytogenetic methods. As such, array-CGH technique may present a suitable complementary test to cytogenetic analysis in cases with a normal karyotype.

A new genus and two new species of Stevardiinae (Characiformes: Characidae) with a hypothesis on their relationships based on morphological and histological data

Lepidocharax, new genus, and Lepidocharax diamantina and L. burnsi new species from eastern Brazil are described herein. Lepidocharax is considered a monophyletic genus of the Stevardiinae and can be distinguished from the other members of this subfamily except Planaltina, Pseudocorynopoma, and Xenurobrycon by having the dorsal-fin origin vertically aligned with the anal-fin origin, vs. dorsal fin origin anterior or posterior to anal-fin origin. Additionally the new genus can be distinguished from those three genera by not having the scales extending over the ventral caudal-fin lobe modified to form the dorsal border of the pheromone pouch organ or to represent a pouch scale in sexually mature males. In this paper, we describe these two recently discovered species and the ultrastructure of their spermatozoa.

Estimating HIV-1 incidence using the serologic testing algorithm for recent HIV infections at HIV counseling and testing centers in the city of São Paulo, Brazil

The network of HIV counseling and testing centers in São Paulo, Brazil is a major source of data used to build epidemiological profiles of the client population. We examined HIV-1 incidence from November 2000 to April 2001, comparing epidemiological and socio-behavioral data of recently-infected individuals with those with long-standing infection. A less sensitive ELISA was employed to identify recent infection. The overall incidence of HIV-1 infection was 0.53/100/year (95% CI: 0.31-0.85/100/year): 0.77/100/year for males (95% CI: 0.42-1.27/100/year) and 0.22/100/ year (95% CI: 0.05-0.59/100/year) for females. Overall HIV-1 prevalence was 3.2% (95% CI: 2.8-3.7%), being 4.0% among males (95% CI: 3.3-4.7%) and 2.1% among females (95% CI: 1.6-2.8%). Recent infections accounted for 15% of the total (95% CI: 10.2-20.8%). Recent infection correlated with being younger and male (p = 0.019). Therefore, recent infection was more common among younger males and older females.

Phylogenetic relationships within the speciose family Characidae (Teleostei: Ostariophysi: Characiformes) based on multilocus analysis and extensive ingroup sampling

Background: With nearly 1,100 species, the fish family Characidae represents more than half of the species of Characiformes, and is a key component of Neotropical freshwater ecosystems. The composition, phylogeny, and classification of Characidae is currently uncertain, despite significant efforts based on analysis of morphological and molecular data. No consensus about the monophyly of this group or its position within the order Characiformes has been reached, challenged by the fact that many key studies to date have non-overlapping taxonomic representation and focus only on subsets of this diversity. Results: In the present study we propose a new definition of the family Characidae and a hypothesis of relationships for the Characiformes based on phylogenetic analysis of DNA sequences of two mitochondrial and three nuclear genes (4,680 base pairs). The sequences were obtained from 211 samples representing 166 genera distributed among all 18 recognized families in the order Characiformes, all 14 recognized subfamilies in the Characidae, plus 56 of the genera so far considered incertae sedis in the Characidae. The phylogeny obtained is robust, with most lineages significantly supported by posterior probabilities in Bayesian analysis, and high bootstrap values from maximum likelihood and parsimony analyses. Conclusion: A monophyletic assemblage strongly supported in all our phylogenetic analysis is herein defined as the Characidae and includes the characiform species lacking a supraorbital bone and with a derived position of the emergence of the hyoid artery from the anterior ceratohyal. To recognize this and several other monophyletic groups within characiforms we propose changes in the limits of several families to facilitate future studies in the Characiformes and particularly the Characidae. This work presents a new phylogenetic framework for a speciose and morphologically diverse group of freshwater fishes of significant ecological and evolutionary importance across the Neotropics and portions of Africa.

Allosteric Control of Gating Mechanisms Revisited: The Large Conductance Ca(2+)-Activated K(+) Channel

Large-conductance Ca(2+)-activated K(+) channels (BK) play a fundamental role in modulating membrane potential in many cell types. The gating of BK channels and its modulation by Ca(2+) and voltage has been the subject of intensive research over almost three decades, yielding several of the most complicated kinetic mechanisms ever proposed. A large number of open and closed states disposed, respectively, in two planes, named tiers, characterize these mechanisms. Transitions between states in the same plane are cooperative and modulated by Ca(2+). Transitions across planes are highly concerted and voltage-dependent. Here we reexamine the validity of the two-tiered hypothesis by restricting attention to the modulation by Ca(2+). Large single channel data sets at five Ca(2+) concentrations were simultaneously analyzed from a Bayesian perspective by using hidden Markov models and Markov-chain Monte Carlo stochastic integration techniques. Our results support a dramatic reduction in model complexity, favoring a simple mechanism derived from the Monod-Wyman-Changeux allosteric model for homotetramers, able to explain the Ca(2+) modulation of the gating process. This model differs from the standard Monod-Wyman-Changeux scheme in that one distinguishes when two Ca(2+) ions are bound to adjacent or diagonal subunits of the tetramer.

Development and mechanical testing of a short intramedullary nail for fixation of femoral rotational osteotomy in cerebral palsy patients

Background: Rotational osteotomy is frequently indicated to correct excessive femoral anteversion in cerebral palsy patients. Angled blade plate is the standard fixation device used when performed in the proximal femur, but extensile exposure is required for plate accommodation. The authors developed a short locked intramedullary nail to be applied percutaneously in the fixation of femoral rotational osteotomies in children with cerebral palsy and evaluated its mechanical properties. Methods: The study was divided into three stages. In the first part, a prototype was designed and made based on radiographic measurements of the femoral medullary canal of ten-year-old patients. In the second, synthetic femoral models based on rapid-prototyping of 3D reconstructed images of patients with cerebral palsy were obtained and were employed to adjust the nail prototype to the morphological changes observed in this disease. In the third, rotational osteotomies were simulated using synthetic femoral models stabilized by the nail and by the AO-ASIF fixed-angle blade plate. Mechanical testing was done comparing both devices in bending-compression and torsion. Results: The authors observed proper adaptation of the nail to normal and morphologically altered femoral models, and during the simulated osteotomies. Stiffness in bending-compression was significantly higher in the group fixed by the plate (388.97 +/- 57.25 N/mm) than in that fixed by the nail (268.26 +/- 38.51 N/mm) as torsional relative stiffness was significantly higher in the group fixed by the plate (1.07 +/- 0.36 Nm/degrees) than by the nail (0.35 +/- 0.13 Nm/degrees). Conclusions: Although the device presented adequate design and dimension to fit into the pediatric femur, mechanical tests indicated that the nail was less stable than the blade plate in bending-compression and torsion. This may be a beneficial property, and it can be attributed to the more flexible fixation found in intramedullary devices.