소개
해부학 실험 데이터는 대체로 이미지로 된 결과물이 많으며, 논문작성시 일반적인 용어가 아닌 전문적이고 해부학적인 용어를 써서 기술해야 하는데,
현재 대체로 텍스트를 기반으로 하는 (한다고 생각했던) AI가 이미지 데이터 분석에 얼마나 활용 가능할지 테스트했음.
진행 방법
해부학 실험 데이터를 효과적으로 AI에게 분석을 하도록 하기 위해서 이미지 분석에 탁월하다는 Google Gemini를 활용하였고, 결과를 챗GPT와 비교하였음.
사용한 프롬프트
너는 치아발생학 분야의 전문학자야. 이 이미지는 journal of dental research에 투고하기 위해 만든 Figure 1이야. 치아발생과정을 다룬 해부학적 데이타인데 저널의 author guideline과 sample paper 등을 참고하여 figure legend를 작성해줘. 학술용어를 활용하여 전문가답게 작성해줘.
위 프롬프트와 함께, 일반적인 이미지가 아닌 전문적인 실험 데이터 이미지를 보여줌.
결과
챗GPT의 결과
Figure 1. Developmental stages of tooth morphogenesis in mouse models. (A-D) Immunohistochemical analysis of tooth development at E13.5, E14.5, E16.5, and P0. In these early stages, specific patterns of protein expression can be observed within the dental tissues. The expression of the protein in A indicates early tooth development, localized to the tooth bud (A, dashed line), while in E16.5 (C) the protein (*), which is indicative of later differentiation events, shows increased expression at the enamel organ. (D) At P0, expression patterns suggest the onset of initial mineralization. (E-H) Dynamic changes in protein expression at postnatal stages P0, P6, and P8. Protein localization is shown in these stages, with arrows indicating areas of interest. Specifically, at P0 (E), immunoreactivity is seen in the basal layer of the developing enamel organ (arrows). By P6 (F), the differentiation process is more pronounced, with notable changes in cellular architecture. At P8 (G), the expression pattern remains robust, but distinct layers in the enamel organ become more defined. (I-N) Further temporal progression of tooth development is shown, with detailed analysis at P8, P4, and P6 (I-N). As indicated in the insets, at later stages of development, the odontoblast (Od) and ameloblast (Am) cells are clearly defined. These cells are involved in the formation of dentin and enamel, respectively. (O-R) Immunofluorescence staining at P11 for the control (O) and K14Cre;Fst mutant (Q). The control shows typical expression of the target protein in the developing enamel organ (O), while the mutant (Q) demonstrates altered expression patterns, suggesting potential disruptions in normal tooth development. Insets show magnified images highlighting changes in cellular architecture and localization. Scale bars in all images represent 50 µm. Note: Abbreviations: Am, ameloblast; Od, odontoblast; Rt, root; Cr, crown; Amb, ambinseral.
Gemini의 결과
Figure 1. Spatiotemporal Expression Pattern of [단백질/유전자 명칭] during Murine Odontogenesis and Functional Validation via Genetic Ablation. (A-D) Coronal sections of murine mandibles at embryonic day (E) 13.5 (A), E14.5 (B), E16.5 (C) and postnatal day (P) 0 (D), immunohistochemically stained to visualize [단백질/유전자 명칭] expression (brown signal). At E13.5, [단백질/유전자 명칭] expression is localized within the dental mesenchyme, demonstrating broad distribution. Subsequent stages show a progressive restriction in expression to the dental epithelium and forming enamel organ, with notable signal accumulation in the developing inner enamel epithelium and stellate reticulum at E16.5 (yellow asterisk, C). By P0, expression is predominately localized in the region around the tooth germ. The red dotted line in A denotes the border between the dental epithelium and mesenchyme, and thus the staining is in mesenchyme at this point. (E-J) Postnatal coronal sections of developing molars at P0 (E), P6 (F), and P8 (G) exhibiting reduced [단백질/유전자 명칭] expression within the enamel organ, localizing to the cervical loop (red arrows). (H-J) are higher magnification insets of the boxed areas in E-G demonstrating progressive diminution of signal in the region and a more discrete localization. (K-L) Sagittal sections of a molar at P6. The red arrows point to the region between the ameloblast layer (Am) and odontoblast layer (Od). (K) shows the overall tooth germ and (L) is an enlarged inset image. (M-N) Coronal sections of a murine developing molar at P4 (M) and P6 (N). (M and N) insets are the areas surrounding the cervical loop at higher magnification. Note the absence of brown immuno-staining indicating the lack of expression in the control mice at these stages. (O-R) Immunofluorescent analysis of sagittal sections of P11 molars of control and K14-Cre;Fst conditional knockout mice stained with DAPI (blue), anti-calbindin (Calb, green), and anti-follistatin (Fst, red) antibodies. (O,Q) are lower magnification images with the areas of interest indicated with dashed white boxes. (P,R) are higher magnification insets corresponding to the dashed boxed areas in O, Q. (P) shows robust expression of Fst and Calb within the ameloblast layer in the control mice. (R) shows that genetic ablation of Fst via K14-Cre;Fst conditional deletion leads to a marked reduction in Fst signal within ameloblast cells (white arrow, R) whilst calbindin expression remains in this tissue. Scale bars: 50 µm (A-G, H-J, K, L, M, N) ; 100 µm (O,Q)
결과와 배운 점
특별히 전문적인 용어를 제공하거나 설명하지 않았음에도 불구하고 챗GPT와 Gemini 모두 놀라운 이미지 분석 결과를 보여주며 이를 전문용어를 사용하여 글로 나타내줌.
챗GPT는 군더더기 없이 요구한 내용만 보여주거나 살짝 대충하는 느낌을 줌.
대체로 Gemini가 더 많은 전문용어를 활용하여 상세하게 기술하며, 빠뜨린 부분도 알아서 기술해주는 꼼꼼함과, 설명을 덧붙여주는 친절함을 보여줌.
아마도 프롬프트 입력시 페르소나 설정과 투고할 저널이름을 알려주었기 때문에 그 분야의 전문용어를 활용하는데 강점을 가지지 않았나 싶으나, 그럼에도 불구하고 학술용 데이터 이미지를 분석하는데 매우 놀라운 능력을 가졌음.
그러나 아무래도 100% 정확하게 잘한 것은 아니며 내용 수정을 통해 보완하여야 함.
처음에는 Gemini에게 실시간 화면공유를 활용하여 작성해 보았는데 오히려 저장 이미지를 제공한 것이 나은 결과를 보였음.
앞으로는 더 다양한 부분에서 활용해 볼 예정임.