Presentation is loading. Please wait.

Presentation is loading. Please wait.

Morphological Identification of P. maniculatus and P. leucopus Mice with Java ABSTRACT Efficient methods of distinguishing sister taxa in areas of sympatry.

Similar presentations


Presentation on theme: "Morphological Identification of P. maniculatus and P. leucopus Mice with Java ABSTRACT Efficient methods of distinguishing sister taxa in areas of sympatry."— Presentation transcript:

1 Morphological Identification of P. maniculatus and P. leucopus Mice with Java ABSTRACT Efficient methods of distinguishing sister taxa in areas of sympatry are important for testing theories of niche partitioning, character displacement, and hybridization. Peromyscus leucopus and P. maniculatus are sympatric in many areas of the eastern United States. Previous methods of morphological identification have not matched the accuracy of genetic analysis. Although genetic analysis may serve as the most accurate speciation technique, it remains costly and time consuming. Computer analysis allows additional rigor in morphological identification, so an algorithm was developed in Java to distinguish these species. The user can import a photo of a Peromyscus mouse, with the full body and tail shown, and run a 4- point analysis of fur color, tail to body ratio, ear to body ratio, and dorsal-ventral tail gradient. Logistic regression is applied to these traits to achieve a species classification. Using a data set of 48 species-identified mice, this program accurately classifies 95% of P. leucopus and 90% of P. maniculatus photos. With additional data and statistical optimization, we believe that these results can be improved. These results are promising because they have lowered the implicit bias in morphological identification. This program has many applications in character displacement studies in areas with and without hybridization. Margo Morton 1, Brian Shelburne 1, Richard Phillips 2 OBJECTIVE To determine whether P. leucopus and P. maniculatus mice could be differentiated by morphological characteristics alone, and to determine whether automated characteristic analysis decreased inherent observer bias in previous morphological identification methods. METHODS RESULTS CONCLUSIONS BACKGROUND The idea for a computer program that could identify species of mice was originally proposed by Professor Richard S. Phillips, who believes that field biologists could benefit from an efficient classification algorithm. P. maniculatus and P. leucopus mice are used widely in biological and ecological research, but are sympatric in many areas of the United States and can be difficult to distinguish even by trained biologists. Currently, biologists must rely on genetic sequencing which is time consuming and expensive. This semester, we sequenced six mice, at a cost of $242. This cost of $40 per mouse would not be practical for large character studies involving hundreds of mice. If mice could be accurately identified based on physical characteristics alone, then an algorithm would be much more practical. According to leading biological literature on mouse identification, P. maniculatus and P. leucopus can be distinguished through several characteristics. These features would be subjective and hard to quantify by a human but objective to a computer, such as coloration patterns on a tail. Research led me to conclude that the best way to identify mice was a 4-point analysis of fur color, tail to body ratio, ear to body ratio, and dorsal-ventral tail gradient. Figure 1: P. maniculatusFigure 2: P. leucopus A dataset was developed from images of mice trapped and identified by Prof. Phillips, as well as mouse images from online databases. The final dataset included 48 species-identified mice. The species test was a sequence-based analysis of a hypervariable region of the mitochondrial 16S rRNA gene. Java was selected to code this software, due to the fact that the final software could be executed and run on any machine. Java AWT and Swing libraries were used in order to develop a straightforward and robust user interface, with functionality that feels familiar to users of graphical editing programs such as Photoshop. User interface functions include the ability to load images into the program, analyze with tools, scale the image, and undo/redo actions. The user can import a photo of a Peromyscus mouse, with the full body and tail shown, and run a 4-point analysis of fur color, tail to body ratio, ear to body ratio, and dorsal- ventral tail gradient. Logistic regression is applied to these traits to achieve a species classification. 1. An area of the fur is averaged to get a fur color value, then a distance formula is used to compute the distance to the average color of P. leucopus and P. maniculatus in 3D Colorspace Color Distance = ((r2 - r1) 2 + (g2 - g1) 2 + (b2 - b1) 2 ) 1/2 2. The tail, ear, and body length are measured in pixels, and divided to get ratios of these values 3. A pixel from the dorsal and ventral part of the tail is selected, and a distance formula is used to get a tail gradient value Figure 3. These blocks of color represent the average fur color of a P. leucopus mouse and a P. maniculatus mouse. A computer can use the distance equation to determine how close a fur color is to these averages. Caballas, Rauleen, Stephanie Fore, and Hyun-Joo Kim. "Regional Model for Identifying Peromyscus Leucopus and Peromyscus Maniculatus." Thesis. Truman State University, n.d. Print. Choate, Jerry R. "Identification and Recent Distribution of White-Footed Mice (Peromyscus) in New England." Journal of Mammalogy 54.1 (1973): 41-49. JSTOR. Web. 27 Mar. 2014. Lindquist, Erin S., Charles F. Aquadro, Deedra McClearn, and Kevin J. McGowan. "Field Identification of the Mice Peromyscus Leucopus Noveboracensis and P. Maniculatus Gracilis in Central NewYork." Canadian Field- Naturalist 117.2 (2003): 184-89. Print. Rich, Stephen M., William Kilpatrick, Jodi L. Shippee, and Kenneth L. Crowell. "Morphological Differentiation and Identification of Peromyscus Leucopus and P. Maniculatus in Northeastern North America." Journal of Mammalogy 77.4 (1996): 985-91. JSTOR. Web. 27 Mar. 2014. I believe this research has shown that computer software can accurately classify P. maniculatus and P. leucopus based on physical characteristics alone. Computer classification, when compared to the morphologic identification methods proposed by Truman, Cabalas and Rich, yields similar accuracy for P. leucopus and vastly increased accuracy for P. maniculatus mice. It has also shown to be faster and less costly than traditional genetic analysis methods. This program could be extended to classify additional species of mice, such as P. gossypinus or P. polionotus given a set of unique morphologic identifiers. Travel funding provided by Wittenberg’s Student Development Board. 1. Wittenberg University, Department of Mathematics and Computer Science, Box 720, Springfield, OH 45501 2. Wittenberg University, Department of Biology, Box 720, Springfield, OH 45501 ACKNOWLEDGEMENTS SOURCES Using a data set of 48 species-identified mice, this program accurately classifies 95% of P. leucopus and 90% of P. maniculatus photos. With additional data and statistical optimization, we believe that these results can be improved. The following logistic regression equation was developed for this dataset: In the future, the algorithm may be modified to dynamically generate logistic regression coefficients, in order to more accurately classify new data sets. These results are promising because they have lowered the implicit bias in morphological identification. This software has many applications in character displacement studies in areas with and without hybridization. Mouse Image SOFTWARE OVERVIEW Analyzing the Image Data Saved in Excel Figure 5: Screencaps demonstrating the use of the software. A photograph is selected, loaded into the program, analyzed, and the results are stored in an excel file. Software available for download at github.com/mortonm FUTURE RESEARCH This software has the potential for improvement, and is available under the GNU General Public License v3.0 for anyone to contribute to the development. Possible improvements to the program include: Automation of mouse classification through feature recognition Dynamic generation of new logistic regression coefficients Demonstration of result reproducibility through Receiver Operating Characteristic (ROC) plots Testing the software with a larger image dataset


Download ppt "Morphological Identification of P. maniculatus and P. leucopus Mice with Java ABSTRACT Efficient methods of distinguishing sister taxa in areas of sympatry."

Similar presentations


Ads by Google