利用深度學習探討台灣蛾類雌雄二型性沿著海拔的演化

Deep learning reveals the evolution of sexual dimorphism in moths along elevational gradients in Taiwan

溫雅涵(Ya-Han Wen)

指導教授 ： 沈聖峰

國立臺灣大學/生命科學院/生態學與演化生物學研究所/碩士(2022年)

https://doi.org/10.6342/NTU202200205

摘要

雌雄二型性的機制從達爾文與華萊士以來就爭論不休，達爾文認為是因為跟性別相關的性擇；華萊士則是認為是跟性別相關的天擇產生雌雄二型性．過去生態學家一直對於鱗翅目的雌雄二型性很有興趣．然而，量化雌雄蟲的形態差異一直都缺乏宏觀與客觀的方法．這裡我們透過深度學習技術—自編碼器（AutoEncoder）來研究台灣蛾類雌雄二型性沿海拔的演化機制．
這個模型可以從資料庫大量的資料訓練模型從較低維（512維）的特徵還原蛾成原本的標本照（256*256*3維），學到構成蛾的外觀特徵後再去處理台灣有雌雄二型性小樣本的蛾類標本照資料（7915筆資料，138個物種），從中萃取出特徵後就可以量化雌雄二型性的差異及之後的探討與海拔之間的關係．接著從此算出雌蟲與雄蟲在不同海拔區間的外觀變異、特徵分布均勻度與特徵獨特性．最後模型能視覺化出雌蟲與雄蟲在海拔實際上的外觀變化．
結果顯示雌雄二型性的差異會隨著海拔上升而增加．而這樣的趨勢是由於雄性沿著海拔梯度的外觀變化較小；雌性沿著海拔梯度的外觀變化較大導致．自編碼器視覺化的結果顯示在海拔梯度上的外觀變化雌雄蟲不同．雌蟲在高海拔顏色變深、紋路變明顯．顏色變深符合熱黑化假說（Thermal melanism hypothesis），紋路變明顯可能因為紋路在雌性較不易受到捕食壓力影響，紋路不改變，但因為顏色變深而變明顯；雄蟲在高海拔亮度維持、前翅的紋路變不明顯，顯示出在捕食壓力隨捕食壓力而減少下，由於天擇效應減少但性擇的顯現．這個結果表示了雌雄二型性在海拔上的演化混合了達爾文的天擇與華萊士性擇的作用，不能以單一機制去解釋．另外這份研究也提供了一個新的方法去研究昆蟲的外觀．

關鍵字

雌雄二型性 ； 深度學習 ； 自編碼器 ； 海拔 ； 形態

並列摘要

The mechanism of sexual dimorphism has been debated since Darwin and Wallace. Darwin believed that it caused by sex-specific sexual selection, while Wallace believed it caused by sex-specific natural selection. Ecologists have been interested in sexual dimorphism in Lepidoptera for a long time. However, few macroscopic and objective methods to quantify the morphological differences between male and female. Here, we use a deep learning technique, AutoEncoder, to study the evolution of sexual dimorphism along the elevation of moths in Taiwan. This model can extract the features from photos, so we can quantify moths’ appearance and calculate the differences between males and females. As a result, we can explore the relationship between sexual dimorphism and altitude. Besides, calculating the appearance difference between males and females at low and high elevational groups can realize how natural selection works in different sex along elevational gradient. Finally, AutoEncoder can reconstruct the average appearance in different sex from each elevational groups. We can see the change of appearance in males and females along elevation.
The results show the degree of sexual dimorphism would increase along elevational gradient. This trend is caused by the fact that males maintain morphological variation at high altitude, while females have less morphological variation at high altitude. The visualization results show that morphological changes along elevational gradient are different from males and females. The darker coloration on the wings of females at high altitude are consistent with the thermal melanism hypothesis, and the more apparent patterns on the wings of females at high altitude are caused by suffering from less predation risk. In the contrast, the disappearance of patterns on the forewings of males at high altitude indicates the influence by sexual sexual due to reduce predation risk. This result revels the evolution of sexual dimorphism along elevational gradient is caused by both sexual selection and natural selection. In addition, this study also provides a novel approach to study insect morphology.

並列關鍵字

Sexual dimorphism ； Deep-learning ； AutoEncoder ； Elevation ； Morphology

參考文獻

Cott, H. B. (1940). Adaptive coloration in animals.

Kettlewell, B. (1973). Evolution of melanism: the study of a recurring necessity.

Endler, J. A. (1978). A predator’s view of animal color patterns. In Evolutionary biology (pp. 319-364). Springer.

Kottler, M. J. (1980). Darwin, Wallace, and the origin of sexual dimorphism. Proceedings of the American Philosophical Society, 124(3), 203-226.

Brown, J. H. (1984). On the relationship between abundance and distribution of species. The American Naturalist, 124(2), 255-279.